Kwenzekeni
UDkt. Munjal Shroff, udokotela wengqondo e-Atlanta, waqala ukufuna umsizi we-AI okhethekile wokunakekelwa kwezempilo lapho umkakhe ethola izimpawu ezingachazeki ngasekupheleni kuka-2025. Wabhalisela i-Healz AI, i-chatbot eklanyelwe ukusiza iziguli ezinomdlavuza zilandelele ukuxilonga, ukwelashwa, kanye nokululama. Uhlelo lokusebenza luncome umbono wesibili ku-MD Anderson Cancer Center, okuholele ekuxilongweni kwe-T-cell lymphoma engavamile. Kukho konke ukwelashwa kwakhe, u-Shroff wasebenzisa i-Healz ukuhlola izinketho zokwelashwa ngamakhemikhali futhi aqonde imininingwane yezokwelapha, ephawula ukuthi iziphakamiso ze-AI zazivame ukuhambisana nezincomo zodokotela bakhe. Ngemuva kwezinyanga zokusetshenziswa kanzima, u-Shroff watshala imali ku-Healz, wajoyina abasunguli u-Igor Demishev no-Alex Plat njengomtshali-zimali. Le nkampani, isekelwa umsunguli we-Recursive Superintelligence u-Alexey Dosovitskiy, ibike ukuthi imali engenayo ephindelelayo yonyaka ifinyelele ku-$100,000 ekuqaleni kukaSepthemba, ngezinga lokugcinwa kwabasebenzisi elingu-70% ngenyanga nenyanga kanye nesizinda sabasebenzisi esiyiziguli ezinomdlavuza ezingama-97%.
Ekwindla ka-2025, umkaDkt. Shroff wabhekana nokuqubuka, ukulunywa, ukukhathala, nama-lymph nodes avuvukele. Ngemva kokubonisana okuningi nokuhlolwa okungaphelele, uShroff wafuna usizo lwedijithali futhi wathola i-Healz AI, i-chatbot eyethulwe ku-beta ngonyaka odlule. Isincomo sohlelo lokusebenza sokufuna umbono wesibili eNyuvesi yaseTexas MD Anderson Cancer Center siholele ekuxilongweni okuqinisekile kwe-T-cell lymphoma.
Ukusebenza kwe-Healz kuhlanganisa ukulandelela izimpawu, ukungenisa amadokhumenti (umsebenzisi oyedwa ulayishe amafayela ezokwelapha angu-1,500), kanye "nemodi yembangela yezimpande" ebuza imibuzo yokulandelela efinyelela kwangu-20 ngaphambi kokunikeza iziphakamiso. Inkundla iphinde isize iziguli ukuthi zithole odokotela be-oncologists, zitolike izikena, futhi ziqondise izinhlelo zokwelapha. U-Shroff wasebenzisa leli thuluzi ukuqhathanisa izinketho zokwelashwa ngamakhemikhali, futhi udokotela wakhe ukuqinisekisile ukuncoma kwe-AI njengokufanelekile.
Abasunguli bokuqalisa, u-Igor Demishev no-Alex Plat, bagcizelele ukuthi i-Healz ayifaki odokotela esikhundleni kodwa isebenza njengensiza eyengeziwe. Babike ukuthi imali engenayo ephindelelayo iphindwe kabili inyanga ngayinye kusukela ngoJuni kuya ku-Agasti, ifinyelela ku-$100,000 emalini engenayo yonyaka ephindelelayo ekuqaleni kukaSepthemba, ngezinga lokugcinwa elingu-70% nokuzibandakanya kwansuku zonke noma kweviki okuvela kubasebenzisi abaningi.
Ngemva kokuhlangenwe nakho kwakhe siqu, u-Shroff watshala imali e-Healz, wajoyina umzuliswano wokuxhasa ngemali ohlanganisa u-Alexey Dosovitskiy, umsunguli we-Recursive Superintelligence, inkampani enenani lamaRandi ayizigidi eziyizinkulungwane ezingu-4.65. Abasunguli badalule izinhlelo zokuqoqwa kwezimali okwengeziwe, nakuba amanani nemininingwane yokufinyelela ebanzi ingadalulwanga.
Imininingwane yomthombo: businessinsider.com ↗
Kungani kubalulekile
I-Healz AI ikhombisa ukuthi amathuluzi e-AI akhethekile angakwengeza kanjani ukumela kwesiguli futhi asheshise izinqumo ezibucayi zezokwelapha, ikakhulukazi kumdlavuza ongandile lapho izindlela zokuxilonga ziyinkimbinkimbi. Ngokuletha umbono wesibili ofika ngesikhathi, i-chatbot ingase inciphise ukubambezeleka kokuxilonga, into eyaziwa ukuthi ithinta imiphumela ku-oncology. Ukukhula kwemali engenayo yokuqala yokuqala kanye nokugcinwa okuphezulu kuphakamisa isifiso semakethe sokusekelwa kwesiguli esiqhutshwa yi-AI, okubonisa ukushintshela ekuzisizeni okunikwe amandla yi-AI ekunakekelweni kwezempilo. Ngaphezu kwalokho, ukuzibandakanya komtshalizimali osezingeni eliphezulu njengo-Alexey Dosovitskiy kugcizelela ukukhula kokuzethemba ezinhlelweni ze-AI ezingaphezu kwamamodeli ezilimi ezinkulu ezijwayelekile, okugqamisa ukusebenza kwezentengiselwano kwezixazululo zesizinda esithile se-AI. Kodwa-ke, le ndaba iphinde igcizelele isidingo sokwengamela komuntu, njengoba abasunguli kanye no-Shroff bevuma ukunemba kwezikhathi ezithile kwe-AI kanye nengcindezi okufanele yengeze, hhayi esikhundleni, odokotela.
Icala libonisa isenzakalo esiphathekayo lapho ithuluzi le-AI libe nomthelela ngokuqondile esinqumweni sezokwelapha esibucayi, okungenzeka lifinyeze isikhathi sokuxilongwa okunembile komdlavuza ongavamile—isimo lapho ukubambezeleka kungase kubeke ukuphila engozini.
Imali engenayo yokuqala ye-Healz nokugcinwa okuphezulu kuphakamisa ukuthi iziguli zizimisele ukukhokhela ukuhamba kwezempilo okushayelwa yi-AI, okubonisa imodeli yebhizinisi esebenzayo yezicelo ezikhethekile ze-AI kwezokwelapha.
Ukubandakanyeka komtshalizimali ovelele we-AI kukhombisa ukuzethemba emandleni okuhweba esizinda esithile se-AI, okungase kukhuthaze inzalo yemali yebhizinisi ekuqalweni kwe-AI okugxile kwezempilo.
Ukuvuma kwabasunguli bemikhawulo ye-AI kuqinisa ukubaluleka kokugcina ukubhekwa kodokotela, ukucatshangelwa okubalulekile kwabalawuli nabahlinzeki bezempilo abahlola ukuhlanganiswa kwe-AI.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
Impossibility results in algorithmic fairness (e.g. Kleinberg et al., Chouldechova) show what?
Ongakubuka ngokulandelayo
I-Healz ihlela ukuqongelela imali okwengeziwe ezinyangeni ezizayo, okunganweba isisekelo sayo sabasebenzisi nezici zomkhiqizo. Izibukeli kufanele ziqaphe ukuthi ukuqalisa kuvikela ubambiswano olubanzi lwezokwelapha, kuhlanganiswe nezinhlelo zamarekhodi ezempilo e-elekthronikhi, noma kuthola ukuvunyelwa okusemthethweni okungavumela ukutholwa okubanzi. Ukulandelela ukusebenza "kwemodi ye-root-cause" kanye nesistimu yememori kuzilungiselelo zomhlaba wangempela kuzoba ukhiye wokuhlola ukwethembeka kwethuluzi nomthelela emiphumeleni yesiguli. Okokugcina, impendulo yomkhakha wezokunakekelwa kwempilo ebanzi—ikakhulukazi ezibhedlela nakubadayisi bomshwalense—izobonisa ukuthi ama-chatbots e-AI angashesha kangakanani ukusuka kumathuluzi asekela isiguli ayi-niche aye ezingxenyeni eziyinhloko zezindlela zokunakekela umdlavuza.
Imijikelezo yesikhathi esizayo yokuqongelela imali ingase inikeze imali edingekayo yokukala umkhiqizo, ukuthobela imithetho, nokuhlanganiswa nezinhlelo zesibhedlela.
Ukubambisana okungase kube khona nezikhungo ezinkulu zomdlavuza noma abashwalense bezempilo kunganweba ukufinyelela kwe-Healz futhi kungase kubeke amazinga embonini ukuze kusekelwe isiguli esisizwa yi-AI.
Ukucutshungulwa kokulawula, ikakhulukazi mayelana nezincomo ze-AI ku-oncology, kungase kuthinte ikhono lesiqalo lokumaketha ithuluzi laso ngale kwesigaba se-beta.
Idatha yemiphumela yomsebenzisi-njengezikhathi zokuxilonga kanye nokubambelela ekwelashweni-izobaluleka ukuze kuqinisekiswe umthelela womtholampilo wethuluzi kanye nokuheha ukutholwa okubanzi.