Isimo sezulu njengesayensi siqhubekela phambili sibonga ngokuthuthuka kobuchwepheshe. Njengamanje, kunezinhlelo eziningi zamakhompyutha ezikwazi ukubikezela ngqo ukuthi lizokuna nini futhi kuphi. Inkampani ye Deepmind ithuthukise ubuhlakani bokuzenzela obukwazi ukubikezela cishe ukuthi izonetha nini futhi kuphi. Le nkampani isebenze nezazi zezulu zase-UK ukudala imodeli engcono yokwenza izibikezelo zesikhashana esifushane kunezinhlelo zamanje.
Kulesi sihloko sizokutshela konke odinga ukukwazi mayelana nesikhwama seRobleda, ubuchwepheshe besimo sezulu esibikezela inkampani DeepMind.
Isimo sezulu
I-DeepMind, inkampani yezobunhloli eyakhiwe eLondon, uyaqhubeka nomsebenzi wakhe wokusebenzisa ukufunda okujulile ezinkingeni ezinzima zesayensi. I-DeepMind isungule ithuluzi lokufunda elijulile elibizwa nge-DGMR ngokubambisana ne-Met Office yaseBritish National Weather Service, engabikezela ngokunembile amathuba emvula emizuzwini engama-90 ezayo. Ingenye yezinselelo ezinzima kakhulu ekubikezeleni isimo sezulu.
Uma kuqhathaniswa namathuluzi akhona, ochwepheshe abaningi bakholelwa ukuthi ukubikezela kwe-DGMR kuhamba phambili ezintweni eziningana, kufaka phakathi ukubikezela kwendawo, ububanzi, ukuhamba namandla emvula, isikhathi esingu-89%. Ithuluzi elisha le-DeepMind livula ukhiye omusha ku-biology ososayensi abakade bezama ukuwuxazulula amashumi eminyaka.
Kodwa-ke, ngisho nokuthuthuka okuncane kokubikezela kubalulekile. Ukubikezela imvula, ikakhulukazi imvula enamandla, kubalulekile ezimbonini eziningi, kusukela emisebenzini yangaphandle kuya kwezinsizakalo zezindiza kanye nezimo eziphuthumayo. Kodwa ukuthola kahle kunzima. Ukunquma ukuthi angakanani amanzi asesibhakabhakeni nokuthi azowela nini futhi kuphi lapho kuya ngezinqubo eziningi zezulu, njengokushintsha kwezinga lokushisa, ukwakheka kwamafu, nomoya. Zonke lezi zinto ziyinkimbinkimbi ngokwazo ngokwazo, kepha ziyinkimbinkimbi kakhulu uma zihlanganiswa.
Ubuchwepheshe obukhona bokuqagela obusetshenziswa kakhulu busebenzisa inani elikhulu lokufaniswa kwamakhompiyutha kwe-physics yomkhathi. Lokhu kufanelekile ekubikezelweni kwesikhathi eside, kepha akukuhle kakhulu ekubikezeleni okuzokwenzeka ehoreni elilandelayo. Lokhu kubizwa ngokuthi isibikezelo ngokushesha.
Ukuthuthukiswa kwe-DeepMind
Izindlela zokufunda ezijulile zangaphambilini zakhiwe, kepha lezi zindlela zivame ukusebenza kahle ngendlela eyodwa, njengokuqagela indawo, nokulimaza enye, njengokuqagela amandla. Idatha yeRadar yemvula enamandla esiza ukubikezela imvula esheshayo ihlala iyinselelo enkulu kwabezimo zezulu.
Ithimba le-DeepMind lisebenzise idatha ye-radar ukuqeqesha i-AI yabo. Amazwe amaningi nezifunda zivame ukushicilela izifinyezo zezilinganiso ze-radar ezilandelela ukwakheka kwamafu nokunyakaza usuku lonke. Isibonelo, e-UK, ukufundwa okusha kuthunyelwa njalo ngemizuzu emihlanu. Ngokubeka lezi zifinyezo ndawonye, ungathola ividiyo eshukumayo yokumisa-ukukhombisa esesikhathini ebonisa ukuthi iphethini lemvula yezwe lishintsha kanjani.
Abaphenyi bathumela le datha kunethiwekhi yezizukulwane ezijulile efana ne-GAN, eyi-AI eqeqeshiwe engakwazi ukukhiqiza amasampula wedatha amasha afana kakhulu nedatha yangempela esetshenzisiwe ekuqeqesheni. IGAN isetshenziselwe ukukhiqiza ubuso obungamanga, kufaka phakathi iRembrandt mbumbulu. Kulokhu, i-DGMR (emele i- "Generative Deep Rain Model") ifunde ukukhiqiza izifinyezo zamanga ze-radar eziqhubeka nokulandelana kwesilinganiso uqobo.
Ukuhlolwa kwe-DeepMind AI
UShakir Mohamed, obehola ucwaningo kwaDeepMind, uthe lokhu kuyefana nokubuka amafreyimu ambalwa ebhayisikobho nokuqagela ukuthi kuzokwenzekani ngokulandelayo. Ukuhlola le ndlela, iqembu licele izazi zezimo zezulu ezingama-56 ezivela e-Bureau of Meteorology (ezazingabandakanyeki emsebenzini) ukuthi zihlole ukulingisa okuphucukile ngokwedlulele neqoqo labaphikisi.
Abantu abangama-89% bathe bathanda imiphumela enikezwe yi-DGMR. Ama-algorithm wokufunda ngomshini ngokuvamile azama ukwenzela isilinganiso esilula sokuthi izibikezelo zakho zihle kanjani. Noma kunjalo, isibikezelo sezulu sinezici eziningi ezahlukene. Mhlawumbe ukubikezela kuthole amandla emvula angalungile endaweni efanele, noma esinye isibikezelo sithole inhlanganisela efanele yamandla kepha endaweni engafanele, njalonjalo.
IDeepMind ithe izokhipha ukwakheka kwawo wonke amaprotheni aziwa yisayensi. Le nkampani isebenzise ubuhlakani bokufakelwa kwamaprotheni bayo be-AlphaFold ukukhiqiza izakhiwo ze-proteome yomuntu, kanye nemvubelo, izimpukane zezithelo namagundane.
Ukusebenzisana phakathi kwe-DeepMind ne-Met Office yisibonelo esihle sokusebenza nabasebenzisi bokugcina ukuqedela ukuthuthukiswa kwe-AI. Ngokusobala lona umqondo omuhle, kepha imvamisa akwenzeki. Ithimba lisebenze kulo msebenzi iminyaka eminingana futhi okuvela kochwepheshe abavela ku-Bureau of Meteorology kwakha iphrojekthi. USuman Ravuri, usosayensi ocwaninga kwa-DeepMind, uthe: "Kukhuthaza ukuthuthukiswa kwemodeli yethu ngendlela ehlukile kunokuqaliswa kwethu." "Ngaphandle kwalokho, besingadala imodeli ebingazusizakala ekugcineni."
I-DeepMind nayo izimisele ukukhombisa ukuthi i-AI yayo inezinhlelo zokusebenza ezisebenzayo. OkweShakir, i-DGMR ne-AlphaFold kuyingxenye yendaba efanayo: inkampani isebenzisa iminyaka yabo yesipiliyoni ukuxazulula amapuzzle. Mhlawumbe isiphetho esibaluleke kunazo zonke lapha ukuthi i-DeepMind ekugcineni isiqalile ukufakwa kuhlu lwezinkinga zesayensi zomhlaba wangempela.
Intuthuko ekubikezelweni kwesimo sezulu
Ukubikezela isimo sezulu kumele kusekelwe ukuthuthukiswa kobuchwepheshe njengoba sisondela futhi sisondela ekuqondeni ngokugcwele ukuthi umoya wethu usebenza kanjani. Izikhathi eziningi umuntu nokubala kwakhe kungaba ngaphansi kwamaphutha ajwayelekile angagwenywa ngokwakhiwa kobuhlakani bokufakelwa.
Ukubikezelwa kwesimo sezulu kuyisihluthulelo sokuba ngumuntu ngoba singasebenzisa okuningi imithombo yamanzi esebenza kahle kakhudlwana futhi ugweme ezinye izinhlekelele ezivunguvungwini nasezimvuleni ezinkulu. Ngalesi sizathu, izazi zezulu ziyavuma ngokuya ngokwengeziwe ukwakha amaphrojekthi obuhlakani bokufakelwa wokubikezela imvula.
Ngiyethemba ukuthi ngalolu lwazi ungafunda kabanzi ngephrojekthi ye-DeepMind nezimpawu zayo.