UWatson akazange amlume udokotela, futhi kahle kakhulu

Nakuba, njengakweminye imikhakha eminingi, umdlandla wokushintsha odokotela esikhundleni se-AI usunciphile ngemva kochungechunge lwezinkinga zokuxilonga, umsebenzi wokuthuthukisa imithi esekelwe ku-AI usaqhubeka. Ngoba, nokho, basanikeza amathuba amahle kanye nethuba lokuthuthukisa ukusebenza kahle kokusebenza ezindaweni eziningi zayo.

I-IBM yamenyezelwa ku-2015, futhi ku-2016 ithole ukufinyelela kwedatha ezinkampanini ezine ezinkulu zedatha yeziguli (1). Okudume kakhulu, ngenxa yemibiko eminingi yemithombo yezindaba, futhi ngesikhathi esifanayo iphrojekthi yokuvelela kakhulu esebenzisa ubuhlakani bokwenziwa obusezingeni eliphezulu obuvela ku-IBM yayihlobene ne-oncology. Ososayensi bebezama ukusebenzisa izinsiza eziningi zedatha ukuze bayicubungule ukuze bayiguqule ibe yimithi yokwelapha ejwayeleke kahle yokulwa nomdlavuza. Inhloso yesikhathi eside bekuwukwenza uWatson abe ngunompempe izivivinyo zokwelashwa nemiphumela njengoba udokotela engenza.

1. Enye yemibono yohlelo lwezokwelapha lwe-Watson Health

Nokho, kwavela ukuthi Watson ayikwazi ukubhekisela ngokuzimela ezincwadini zezokwelapha, futhi ayikwazi ukukhipha ulwazi kumarekhodi ezokwelapha eziguli. Nokho, icala elibi kakhulu ayebekwe lona laliwukuthi ukwehluleka ukuqhathanisa ngokuphumelelayo isiguli esisha nezinye iziguli ezinomdlavuza ezindala futhi zibone izimpawu ezingabonakali ekuqaleni.

Kuyavunywa, kwakukhona odokotela abathile be-oncologist abathi bayasethemba isahlulelo sakhe, nakuba ikakhulukazi ngokweziphakamiso zikaWatson zokwelashwa okujwayelekile noma njengombono owengeziwe, owengeziwe wezokwelapha. Abaningi baveze ukuthi lolu hlelo luzoba umtapo wolwazi ozenzakalelayo wodokotela.

Njengomphumela wokubuyekezwa okungathopha kakhulu okuvela ku-IBM izinkinga ngokuthengiswa kohlelo lwe-Watson ezikhungweni zezokwelapha zase-US. Abamele abadayisi be-IBM bakwazile ukuyidayisela ezinye izibhedlela e-India, South Korea, Thailand nakwamanye amazwe. E-India, odokotela () bahlole izincomo zikaWatson ezimweni ezingama-638 zomdlavuza webele. Izinga lokuthobela izincomo zokwelashwa ngama-73%. Okubi kakhulu Watson uphume eGachon Medical Center eSouth Korea, lapho izincomo zakhe ezihamba phambili zeziguli ezinomdlavuza we-colorectal ezingama-656 zifana nezincomo zochwepheshe ezingamaphesenti angama-49 kuphela wesikhathi. Odokotela bakuhlolile lokho U-Watson akenzanga kahle ngeziguli ezindalangokwehluleka ukubanikeza imishanguzo ethile evamile, futhi benza iphutha elibucayi lokuqapha ukwelashwa okunonya kwezinye iziguli ezine-metastatic disease.

Ekugcineni, nakuba umsebenzi wakhe njengodokotela wokuxilonga nodokotela ubhekwa njengengaphumeleli, kunezindawo aye wabonakala ewusizo kakhulu kuzo. Umkhiqizo Watson for Genomics, eyasungulwa ngokubambisana neNyuvesi yaseNorth Carolina, iYunivesithi yaseYale, nezinye izikhungo, isetshenziswa amalabhorethri ofuzo ukulungiselela imibiko yama-oncologists. I-Watson ilanda ifayela lohlu ukuguqulwa kofuzo esigulini futhi angakwazi ukukhiqiza umbiko ngamaminithi ahlanganisa iziphakamiso zayo yonke imithi ebalulekile nokuhlolwa komtholampilo. U-Watson uphatha ulwazi lofuzo kalulangenxa yokuthi ethulwa ngamafayela ahlelekile futhi awaqukethe izinto ezingaqondakali - noma kukhona ukuguqulwa noma akukho ukuguqulwa.

Abalingani be-IBM eNyuvesi yaseNorth Carolina bashicilele iphepha mayelana nokusebenza kahle ngo-2017. U-Watson uthole ukuguqulwa okungaba okubalulekile okungazange kukhonjwe izifundo zabantu kuma-32% azo. iziguli zafunda, okuzenza abantu abakulungele ukuthola umuthi omusha. Nokho, abukho ubufakazi bokuthi ukusetshenziswa kuholela emiphumeleni engcono yokwelapha.

Ukugcinwa kwamaprotheni ekhaya

Lesi kanye nezinye izibonelo eziningi zifaka isandla enkolelweni ekhulayo yokuthi konke ukushoda ekunakekelweni kwezempilo kuyalungiswa, kodwa sidinga ukubheka izindawo lapho lokhu kungasiza ngempela, ngoba abantu abenzi kahle kakhulu lapho. Insimu enjalo, ngokwesibonelo, ucwaningo lwamaprotheni. Ngonyaka odlule, kwavela ulwazi lokuthi lungabikezela ngokunembile ukuma kwamaprotheni ngokulandela ukulandelana kwawo (2). Lona umsebenzi wendabuko, ngaphezu kwamandla hhayi abantu kuphela, kodwa ngisho namakhompyutha anamandla. Uma sifunda kahle ukumodela okunembayo kokusonteka kwamangqamuzana amaprotheni, kuzoba namathuba amakhulu okwelashwa kwezakhi zofuzo. Ososayensi banethemba lokuthi ngosizo lwe-AlphaFold sizofunda imisebenzi yezinkulungwane, futhi lokhu kuzosivumela ukuba siqonde izimbangela zezifo eziningi.

Umfanekiso 2. Ukusonteka kwamaprotheni okufaniswe ne-DeepMind's AlphaFold.

Manje sazi amaprotheni ayizigidi ezingamakhulu amabili, kodwa sikuqonda ngokugcwele ukwakheka nokusebenza kwengxenye encane yazo. Amaprotheni iyisakhiwo esiyisisekelo sezinto eziphilayo. Banesibopho sezinqubo eziningi ezenzeka kumaseli. Ukuthi basebenza kanjani nokuthi benzani kunqunywa ukwakheka kwabo kwe-50D. Bathatha ifomu elifanele ngaphandle kwemiyalo, beqondiswa imithetho ye-physics. Sekungamashumi eminyaka, izindlela zokuhlola zibe yindlela eyinhloko yokunquma ukuma kwamaprotheni. Ngama-XNUMXs, ukusetshenziswa Izindlela ze-X-ray crystallographic. Kule minyaka eyishumi edlule, seliyithuluzi lokucwaninga elikhethwayo. i-crystal microscopy. Ngama-80s kanye nama-90s, kwaqala umsebenzi wokusebenzisa amakhompiyutha ukuze kutholwe ukuma kwamaphrotheni. Nokho, imiphumela ayizange ibanelise ososayensi. Izindlela ezazisebenza kwamanye amaprotheni azizange zisebenze kwabanye.

Kakade ngo-2018 I-AlphaFold uthole ukuqashelwa ochwepheshe ku amaprotheni modeling. Nokho, ngaleso sikhathi yasebenzisa izindlela ezicishe zifane nezinye izinhlelo. Ososayensi bashintsha amaqhinga futhi benza elinye, eliphinde lasebenzisa ulwazi mayelana nemikhawulo engokomzimba neyejiyomethri ekugoqweni kwama-molecule amaprotheni. I-AlphaFold unikeze imiphumela engalingani. Ngezinye izikhathi wenza kangcono, ngezinye izikhathi ezimbi kakhulu. Kodwa cishe izingxenye ezimbili kwezintathu zokubikezela kwakhe zaqondana nemiphumela etholwe ngezindlela zokuhlola. Ekuqaleni konyaka wesi-2, i-algorithm ichaze ukwakheka kwamaprotheni amaningana egciwane le-SARS-CoV-3. Kamuva, kwatholakala ukuthi izibikezelo zephrotheni ye-Orf2020a ziyahambisana nemiphumela etholwe ngokuhlolwa.

Akukhona nje kuphela ngokutadisha izindlela zangaphakathi zokugoqa amaprotheni, kodwa futhi mayelana nokuklama. Kusetshenziswe abacwaningi bohlelo lwe-NIH BRAIN ukufunda ngomshini thuthukisa iphrotheni ekwazi ukulandelela amazinga e-serotonin yobuchopho ngesikhathi sangempela. I-Serotonin iyi-neurochemical edlala indima ebalulekile endleleni ubuchopho obulawula ngayo imicabango nemizwa yethu. Isibonelo, ama-antidepressants amaningi aklanyelwe ukushintsha izimpawu ze-serotonin ezidluliselwa phakathi kwama-neurons. Esihlokweni esikumagazini i-Cell, ososayensi bachaza indlela abasebenzisa ngayo i-Advanced izindlela zobunjiniyela bezakhi zofuzo guqula iphrotheni yebhaktheriya ibe yithuluzi lokucwaninga elisha elingasiza ukulandelela ukudluliswa kwe-serotonin ngokunemba okukhulu kunezindlela zamanje. Ukuhlolwa kwangaphambilini, ikakhulukazi kumagundane, kubonise ukuthi inzwa ingakwazi ukubona ngokushesha izinguquko ezicashile kumazinga e-serotonin yobuchopho ngesikhathi sokulala, ukwesaba nokusebenzelana komphakathi, futhi ihlole ukusebenza kwezidakamizwa ezintsha ezithinta ingqondo.

Ukulwa nobhubhane akuzange kube yimpumelelo ngaso sonke isikhathi

Phela lolu bekuwubhubhane lokuqala esabhala ngalo kwaMT. Kodwa-ke, ngokwesibonelo, uma sikhuluma ngayo kanye inqubo yokukhula kwalolu bhubhane, lapho-ke esigabeni sokuqala, i-AI yayibonakala iyiphutha. Izifundiswa ziye zakhononda ngalokho ukuhlakanipha okungekhona okwangempela ayikwazi ukubikezela kahle izinga lokusabalala kwe-coronavirus ngokusekelwe kudatha yezifo eziwubhubhane zangaphambilini. “Lezi zixazululo zisebenza kahle kwezinye izindawo, njengokuqaphela ubuso obunenani elithile lamehlo nezindlebe. Ubhubhane lwe-SARS-CoV-2 Lezi izehlakalo ezazingaziwa ngaphambili kanye nezinto eziningi eziguquguqukayo ezintsha, ngakho ubuhlakani bokwenziwa obusekelwe kudatha yomlando eyasetshenziswa ukuyiqeqesha abusebenzi kahle. Lo bhubhane ukhombisile ukuthi kudingeka sibheke obunye ubuchwepheshe nezindlela, ”kusho uMaxim Fedorov waseSkoltech esitatimendeni sika-Ephreli 2020 kwabezindaba baseRussia.

Ngokuhamba kwesikhathi kwaba khona Nokho ama-algorithms abonakala efakazela ukuba wusizo okukhulu kwe-AI ekulweni ne-COVID-19. Ososayensi e-US bathuthukise uhlelo ekwindla ka-2020 lokubona izici zokukhwehlela kubantu abane-COVID-19, noma ngabe babengenazo ezinye izimpawu.

Lapho imithi yokugoma ivela, kwavela umqondo wokusiza ukugoma abantu. Angakwazi, isibonelo ukusiza imodeli yokuthutha kanye nokusetshenziswa kwemithi yokugoma. Futhi ekunqumeni ukuthi yiziphi imiphakathi okufanele zigonywe kuqala ukuze kubhekwane nalolu bhubhane ngokushesha. Kungasiza futhi ukubikezela isidingo kanye nokwandisa isikhathi nesivinini sokugoma ngokuhlonza ngokushesha izinkinga kanye nezingqinamba ku-logistics. Inhlanganisela yama-algorithms anokuqapha okuqhubekayo futhi inganikeza ngokushesha ulwazi mayelana nemiphumela engemihle engenzeka kanye nemicimbi yezempilo.

lezi amasistimu asebenzisa i-AI ekwandiseni nasekuthuthukiseni ukunakekelwa kwezempilo sekuvele kwaziwa. Izinzuzo zabo ezingokoqobo zaziswa; isibonelo, uhlelo lokunakekelwa kwezempilo olwakhiwe yi-Macro-Eyes e-Stanford University e-US. Njengoba kwenzeka kwezinye izikhungo zezokwelapha eziningi, inkinga kwaba ukushoda kweziguli ezingafikanga lapho ziyoqokwa khona. Amehlo Amakhulu yakhe uhlelo olungabikezela ngokuthembekile ukuthi yiziphi iziguli okungenzeka zibe khona. Kwezinye izimo, angase futhi aphakamise ezinye izikhathi nezindawo zemitholampilo, okungase kwandise amathuba okuthi isiguli sivele. Kamuva, ubuchwepheshe obufanayo basetshenziswa ezindaweni ezihlukahlukene ukusuka e-Arkansas kuya eNigeria ngokusekelwa, ikakhulukazi, i-US Agency for International Development i.

ETanzania, iMacro-Eyes isebenze kuphrojekthi okuhloswe ngayo ukwandisa amazinga okugonywa kwezingane. Isofthiwe ihlaziye ukuthi mingaki imithamo yemithi yokugoma okudingeka ithunyelwe esikhungweni sokugoma esinikeziwe. Wakwazi nokuhlola ukuthi yimiphi imindeni engase ibe manqikanqika ukugoma izingane zayo, kodwa ingathonywa ngezimpikiswano ezifanele kanye nendawo yesikhungo sokugoma endaweni ekahle. Ngokusebenzisa le software, uhulumeni waseTanzania ukwazile ukwandisa ukusebenza kahle kohlelo lwakhe lokugoma ngama-96%. futhi yehlise imfucuza yokugoma ibe ngu-2,42 kubantu abayi-100.

E-Sierra Leone, lapho idatha yezempilo yabahlali yayingekho, inkampani yazama ukufanisa lokhu nolwazi olumayelana nemfundo. Kuvele ukuthi isibalo sothisha nabafundi besodwa sanele ukubikezela amaphesenti angama-70. ukunemba kokuthi umtholampilo wendawo uyawathola yini amanzi ahlanzekile, osekuyinkomba yolwazi mayelana nempilo yabantu abahlala khona (3).

3. Umdwebo we-Macro-Eyes wezinhlelo zokunakekelwa kwezempilo eziqhutshwa yi-AI e-Afrika.

Inganekwane kadokotela womshini ayipheli

Naphezu kokwehluleka Watson zisathuthukiswa izindlela ezintsha zokuxilonga futhi zithathwa njengezithuthuke kakhulu. Ukuqhathanisa okwenziwe eSweden ngoSepthemba 2020. esetshenziswa ekuxilongeni umdlavuza webele wabonisa ukuthi okungcono kakhulu kubo kusebenza ngendlela efanayo ne-radiologist. Ama-algorithms ahlolwe kusetshenziswa izithombe ezicishe zibe izinkulungwane eziyisishiyagalolunye ze-mammography ezitholwe ngesikhathi sokuhlolwa okujwayelekile. Amasistimu amathathu, aqokwe njenge-AI-1, AI-2 kanye ne-AI-3, azuze ukunemba okungu-81,9%, 67%. kanye no-67,4%. Uma kuqhathaniswa, kuma-radiologists ahumusha lezi zithombe njengezokuqala, lesi sibalo sasingu-77,4%, futhi esimweni ama-radiologistsobe ngowesibili ukuyichaza, ibingamaphesenti angama-80,1. Ama-algorithms angcono kakhulu akwazile nokuthola amacala izazi ze-radiologist eziphuthelwe ngesikhathi sokuhlolwa, futhi abesifazane batholakala begula esikhathini esingaphansi konyaka.

Ngokwabacwaningi, le miphumela iyakufakazela lokho ama-algorithms obuhlakani bokwenziwa ukusiza ukulungisa ukuxilonga okungelona iqiniso okwenziwa ama-radiologists. Ukuhlanganisa amandla e-AI-1 nodokotela oyi-radiologist omaphakathi kwenyuse isibalo sabatholakele abanomdlavuza wamabele ngo-8%. Ithimba laseRoyal Institute elenza lolu cwaningo lilindele ukuthi ikhwalithi ye-AI algorithms iqhubeke nokukhula. Incazelo egcwele yocwaningo ishicilelwe ku-JAMA Oncology.

W esikalini samaphuzu amahlanu. Njengamanje, sibona ukusheshisa okuphawulekayo kobuchwepheshe futhi sifinyelela izinga le-IV (okuzenzakalelayo okuphezulu), lapho isistimu icubungula ngokuzenzakalelayo idatha etholiwe futhi ihlinzeka uchwepheshe ngolwazi oluhlaziywe ngaphambilini. Lokhu konga isikhathi, kugwema iphutha lomuntu futhi kunikeza ukunakekelwa kwesiguli okuphumelelayo. Yilokho ahlulela ngakho ezinyangeni ezimbalwa ezedlule UStan A.I. emkhakheni wezokwelapha eduze kwakhe, u-prof. Janusz Braziewicz kusukela ku-Polish Society for Nuclear Medicine esitatimendeni esiya ku-Polish Press Agency.

4. Ukubuka ngomshini izithombe zezokwelapha

Ama-algorithms, ngokusho kochwepheshe abanjengo-prof. Brazievichesibalulekile kulo mkhakha. Isizathu siwukunyuka ngokushesha kwenani lokuhlolwa kwezithombe zokuxilonga. Kuphela isikhathi 2000-2010. inani lokuhlolwa kwe-MRI nokuhlolwa liye landa ngokuphindwe kashumi. Ngeshwa, inani lodokotela abangochwepheshe abatholakalayo abangawenza ngokushesha nangokuthembekile alikanyuki. Kukhona nokushoda kochwepheshe abaqeqeshiwe. Ukusetshenziswa kwama-algorithms asekelwe ku-AI konga isikhathi futhi kuvumela ukumiswa okuphelele kwezinqubo, kanye nokugwema iphutha lomuntu kanye nokwelashwa okuphumelelayo, okuqondene nomuntu ezigulini.

Njengoba kwavela, futhi i-forensic Medicine angazuza kukho ukuthuthukiswa kobuhlakani bokwenziwa. Ongoti balo mkhakha bangakwazi ukunquma isikhathi esiqondile sokufa komuntu oshonile ngokuhlaziywa kwamakhemikhali okuphuma kwezibungu nezinye izidalwa ezidla izicubu ezifile. Inkinga iphakama lapho inhlanganisela yezimfihlo ezivela ezinhlotsheni ezahlukene ze-necrophage ifakiwe ekuhlaziyeni. Yilapho ukufunda komshini kuqala khona. Ososayensi baseNyuvesi yase-Albany sebethuthukile indlela yobuhlakani bokwenziwa evumela ukuhlonzwa ngokushesha kwezinhlobo zezikelemu ngokusekelwe "kwezigxivizo zeminwe zamakhemikhali". Iqembu laqeqesha uhlelo lwalo lwekhompiyutha lisebenzisa inhlanganisela yenhlanganisela yamakhemikhali ahlukahlukene avela ezinhlotsheni eziyisithupha zezimpukane. Uthole amasiginesha amakhemikhali ezibungu zezinambuzane esebenzisa i-mass spectrometry, ekhomba amakhemikhali ngokulinganisa ngokunembile isilinganiso sobunzima neshaji kagesi ye-ion.

Ngakho-ke, njengoba ubona, noma kunjalo I-AI njengomseshi ophenyayo ayinhle kakhulu, ingaba wusizo kakhulu elebhu ye-forensic. Mhlawumbe besilindele lukhulu kuyena ngalesi sikhathi, silindele ama-algorithms azokhipha odokotela emsebenzini (5). Uma sibheka ukuhlakanipha okungekhona okwangempela ngokweqiniso, egxile ezinzuzweni ezithile ezingokoqobo kunokujwayelekile, umsebenzi wakhe wezokwelapha ubukeka uthembisa kakhulu futhi.

5. Umbono wemoto kadokotela