Isifinyezo: Abacwaningi bathuthukise i-AI-powered speech neuroprosthesis ehlukanisa amasiginali ahlanganayo aphezulu abe yinkulumo yokwenziwa eshelelayo, ethembekile. Ipayipi lokufunda okujulile lezigaba ezimbili lisebenzisa isendlalelo sokuqoshwa kwefonetiki esibhangqwe nemodeli yolimi yokucubungula ulimi ukuze kuzuzwe ukunemba kwamagama okungaphezu kuka-99% nokubambezeleka kokucubungula okungama-millisecond angu-30 nje.
Ihlolwe kumuntu one-ALS ethuthukisiwe, isixhumi esibonakalayo sinikeze amandla ukuvezwa kwesikhathi sangempela kwamagama ayizigidi ezingu-2.7 phakathi neminyaka emibili, okuhlanganisa nokuguquguquka kwephimbo nokucula, okumaka ingqopha-mlando yokuguqula ukuxhumana kobuchopho nekhompyutha.
Amaqiniso Abalulekile
- Ukuphula Isithiyo Sokulayishwa Kwedatha: Amalungu afanayo esimanjemanje athwebula idatha yokudubula evela kumakhulu ama-neurons angawodwana kanyekanye. Ngenkathi izibalo zezibalo zendabuko ziwa ngokuzama ukwenza umqondo walesi sikhukhula sedatha esikhulu ngesikhathi sangempela, uDkt. Stavisky usebenzise amamodeli e-AI athuthukisiwe ukuze afunde, ahlele, futhi anqume la maphethini emizwa ayinkimbinkimbi ngokushesha.
- I-Dual-Stage AI Decoding Architecture: I-neuroprosthesis izuza ukuhunyushwa kwenkulumo yemvelo ngokuqondisa idatha ye-cortical eluhlaza ngokusebenzisa ipayipi elikhethekile, elinezinyathelo ezimbili lokufunda okujulile:
- Isendlalelo Sefonetiki: Isendlalelo sokuqala se-AI sinquma umsebenzi wobuchopho wesikhathi sangempela sibe amafonemu—amabhlogo omsindo ayisisekelo akha inkulumo.
- Isendlalelo Solimi: Isendlalelo sesibili sisebenzisa izakhiwo zemodeli yolimi olukhulu (LLM) ukuhlela ngokushesha lawo mazwi abe amagama ahlanzekile, imishwana, nemisho ehambisanayo.
- I-Real-Time Voice Synthesis & Iphimbo: Kunokuba ikhiphe umbhalo oyisicaba esikrinini sekhompuyutha, isixhumi esibonakalayo sisebenzisa imodeli yokufunda ejulile eqonde ezwini. Lokhu kuvumela umhlanganyeli ukuthi alawule izwi elizwakalayo, elenziwe ngendlela eqondile kumarekhodi akhe angaphambi kwe-ALS. Ukubambezeleka kokucubungula kuphansi kakhulu (ama-millisecond angu-30) kangangokuthi kufana nengxoxo yomuntu yemvelo, okuvumela umsebenzisi ukushintsha ithoni, ukulungisa iphimbo, ngisho nokucula.
- Umthelela Ongokoqobo Oshintsha Impilo: Ngaphezulu kwewindi leminyaka emibili lokuthunyelwa kwekhaya, umhlanganyeli ukhiqize ngempumelelo 2.7 million amagama ngokubonisana kwengqondo kuphela. I-BCI yomtholampilo inikeze ukuzimela okuphelele kokuxhumana, ivumela umsebenzisi ukuthi ahlanganyele ezingxoxweni ezinothile zansuku zonke nomndeni, alawule ngokuzimela ikhompuyutha yomuntu siqu, futhi alondoloze umsebenzi wesikhathi esigcwele.
- I-Focus Shift Eqhutshwa Isiguli: Ekuqaleni komsebenzi wakhe wezemfundo, uDkt. Stavisky wagxila kakhulu kuma-BCI asekelwe ezimotweni okulawula izingalo zerobhothi. Kodwa-ke, impendulo engaguquki evela ezigulini ezikhubazekile yashintsha indlela yakhe yokuphila. Ngenkathi izinto ezihambayo ziwusizo, iziguli zathi ukubuyisela izwi lazo kwakuyinto ebaluleke kakhulu, okwenza kwaba nethonya eliphakathi nendawo ku-neuroscience yokukhuluma njengoba nje ubuchwepheshe bokufunda ngomshini wabathengi buqala ukushesha ngo-2018.
- Umbono Wesikhathi Eside: Umgomo oyinhloko kaDkt. Stavisky ubunjiniyela bezwi elingenasici “lokuthatha indawo okuthembekile okuphezulu”, ithuluzi lomtholampilo elithuthuke kakhulu kangangokuthi uma umsebenzisi ekhuluma ngocingo oluvamile, umlaleli ngeke akwazi ngokuphelele ukusho ukuthi izwi lakhiwe. Intuthuko yesikhathi esizayo izogxila ekwehliseni isistimu ibe isifakelo sezokwelapha esingenantambo ngokuphelele, sangaphakathi ngokuphelele kuyilapho kwandiswa ukufinyelela ezigulini ezilawula i-stroke-induced aphasia noma i-cerebral palsy.
Umthombo: I-AAAS
Lapho u-Sergey Stavisky eqala ukucabanga nge-brain-computer interfaces (BCI) njengomuntu oneziqu eBrown University, wagqugquzelwa izici ezintathu. “Ngangithanda ukwakha izinto,” ekhumbula, “futhi ngangifuna ukwenza okuthile kwezokwelapha, kodwa futhi ingqondo yayingihlaba umxhwele.
Leyo nhlanganisela izoholela u-Stavisky ensimini manje esichaza kabusha ngokushesha ukuthi kusho ukuthini ukulahlekelwa, futhi okungenzeka ukuthi uphinde uthole, izwi.
Namuhla, u-Stavisky unguprofesa ohlangene wokuhlinzwa kwezinzwa e-University of California, e-Davis, futhi ungumuntu ohamba phambili ekuthuthukisweni kwe-AI-powered speech neuroprostheses. Umsebenzi wakhe, owaziwa kulo nyaka yi-Chen Institute kanye Isayensi Umklomelo Wokucwaninga Okusheshisiwe kwe-Al, uhlezi ezimpambanweni ze-neuroscience, ukunakekelwa komtholampilo nokufunda ngomshini. Kodwa umnyombo wawo umgomo olula: ukubuyisela ikhono lokukhuluma nabantu abalahlekelwe.
Lowo mgomo uba sobala endabeni yomunye umhlanganyeli ocwaningweni lweqembu lakhe, indoda ephila ne-amyotrophic lateral sclerosis (ALS) eyayingasakwazi ukukhuluma ngendlela eqondakalayo.
Ngedivayisi efakelwayo kanye nohlu lwamamodeli e-AI aqeqeshelwe umsebenzi wakhe wobuchopho owaklanywa u-Stavisky nethimba lakhe, indoda manje isiyakwazi ukukhiqiza imisho eqephuzayo – okokuqala njengombhalo, kwase kuba njengenkulumo yokwenziwa efaniswe ezwini lakhe langaphambi kwe-ALS. Ezikhathini zokusetshenziswa kwansuku zonke, ukhiqize izigidi zamagama.
Ukubhekana nokugcwala kwedatha
Isayensi ngemuva kwaleyo mpumelelo incike eqinisweni eliguqule isayensi yezinzwa kule minyaka eyishumi edlule: ukugcwala ngokweqile kwedatha. “Izimpawu zobuchopho ziyinkimbinkimbi ngempela,” kuchaza uStavisky.
Lapho abacwaningi bake baqopha khona ngama-neuron eyodwa, amasistimu esimanje angakwazi ukuthwebula amasignali kumakhulu ama-neurons ngesikhathi. Kodwa ukuziphatha abazama ukukunquma, njengenkulumo, kuphakathi kwezenzo zomuntu eziyinkimbinkimbi kakhulu, nezindlela zendabuko zezibalo zokucubungula idatha, kusho u-Stavisky, zivele zephule ngaphansi kobunzima obunjalo.
Yena nozakwabo babedinga indlela yokucubungula amanani amakhulu edatha ye-neural ngokushesha okukhulu nangokuvumelana nezimo. “I-AI ibonakale inamandla ahlukile kulokho,” esho.
Kusistimu ka-Stavisky, imodeli eyodwa ihlukanisa umsebenzi wobuchopho ube amafonemu, amayunithi ayisisekelo omsindo olimi. Enye imodeli, edweba izindlela zokumodela zolimi ezinkulu, iguqula lezo zingcingo zibe amagama nemisho. Ngenye indlela yokubhekana nezwi, amasistimu okufunda ajulile akha kabusha imisindo yenkulumo, akhiqize izwi lokwenziwa ngesikhathi sangempela.
Umphumela uwuhlelo olungahumushela inhloso enkulumweni ngokwethembeka, ikhiqize imisindo ezwakalayo ngokulibaziseka okufushane okungama-millisecond angu-30. Lokhu kuyashesha ngokwanele ukulinganisa ingxoxo yemvelo.
Ocwaningweni olushicilelwe ku Umuthi Wemvelo ngo-June, u-Stavisky nozakwabo bachaza ukuthi i-BCI yabo yasiza kanjani umhlanganyeli nge-ALS ukuthi agcine ukuxhumana okucebile phakathi kwabantu nomndeni nabangane bakhe ekhaya, ukulawula ngokuzimela ikhompuyutha yakhe yomuntu siqu, kanye nokusekela umsebenzi wesikhathi esigcwele.
“U-Stavisky wenze i-AI-based neuroprosthesis yenkulumo enomthelela osheshayo futhi oguqulayo,” kusho u-Yury V. Suleymanov, umhleli omkhulu ku- Isayensi. “Kubuyisele ukuxhumana esigulini esikhubazekile esine-amyotrophic lateral sclerosis enamagama anembile angaphezu kuka-99%, okwenza isiguli sikwazi ukuveza amagama ayizigidi ezingu-2.7 phakathi neminyaka emibili sisebenzisa izimpawu zobuchopho kuphela.
Ukusuka ekunyakazeni kuya enkulumeni
U-Stavisky uthe ekuqaleni komsebenzi wakhe, ngenkathi esebenza kuma-BCIs wokunyakaza, kwamholela ekubeni agxile kuma-BCI ukuze akhulume. Wayebone okuthile okungaguquki kuzo zonke iziguli: ukubuyisela ikhono lokuhambisa ikhesa noma ingalo yerobhothi kwakubalulekile, kodwa ukubuyisela ukuxhumana kwakuhlale kuphuthuma kakhulu. “Ukuxhumana bekuhlezi kuyiyona nto ehamba phambili,” esho.
Lokho kuqaphela, kuhlanganiswe nentuthuko evelayo ekufundeni komshini kanye nobuchwepheshe bokuqopha obungaphakathi kwekhohlo, kwamholela ekushintsheni okuphakathi nendawo kusukela ku-motor prosthetics kuya enkulumweni. Ngaleso sikhathi, ukuqoshwa kwenkulumo okuvela kuzimpawu zobuchopho kwakubhekwa kabanzi njengenye yezinkinga ezinzima kakhulu ku-neuroprosthetics. Kepha inqubekelaphambili ku-AI ibikhula ngesivinini ngesikhathi esifanele. Ngisho nezinhlelo zokubizela abathengi zase ziqala ukufinyelela emazingeni okusebenza asebenzisekayo cishe ngo-2018, esho.
Ebheka phambili, u-Stavisky uthe umgomo wesikhathi eside “izwi lokumela elithembekile”—uhlelo olungokwemvelo kangangokuthi uma othile ekhuluma ocingweni, “wawungeke usho ukuthi akulona izwi lakhe lemvelo.” Ikusasa cishe lizobandakanya amadivaysi amancane, afakwe ngokugcwele, futhi angabonakali kakhulu kunezinhlelo zanamuhla zocwaningo. Kuzodinga futhi ukusuka kuma-prototypes aselabhorethri kuye kumathuluzi omtholampilo atholakala kabanzi.
Kakade, insimu iyakhula. Izinkampani seziqala ukungena ezivivinyweni zomtholampilo zama-BCI enkulumo, futhi amalebhu ezemfundo ahlola ukuthi izindlela ezifanayo zingasiza yini abantu abane-stroke-induced aphasia, i-cerebral palsy noma ezinye iziyaluyalu zolimi. Imiphumela, u-Stavisky waphakamisa, ingadlulela ngalé kokukhubazeka.
“Eminyakeni eyishumi edlule, mina noTianqiao sasungula i-Chen Institute ukuze siphishekele umbuzo obalulekile: ubuchopho buveza kanjani ubuhlakani?” kusho uChrissy Luo, umsunguli we-Chen Institute.
“Ngaleso sikhathi sasingeke sicabange ukuthi i-AI yayizoshintsha hhayi nje indlela esifunda ngayo ubuchopho, kodwa lokho esingakufunda kukho.” Ucwaningo lukaDkt.
“Lo mklomelo udalelwe lolu hlobo lomsebenzi kanye, futhi siyaziqhenya ukugubha impumelelo yakhe kanye nozakwethu kwa-AAAS kanye Isayensi. Sizibophezele ekusekeleni abacwaningi abachaza kabusha lokho isayensi engakufeza.”
Imibuzo Ebalulekile Iyaphendulwa:
A: Ukuhambisa ingalo yerobhothi noma ikhesa yekhompuyutha kuncike emiyalweni elula uma kuqhathaniswa, yendawo, ingqondo empeleni icabanga ukuhambisa “phezulu, phansi, kwesokunxele, noma kwesokudla” eduze kwegridi eyisicaba. Inkulumo, nokho, ngokungangabazeki iyisenzo esiyinkimbinkimbi kunazo zonke ezenziwa abantu. Ukuze kuzwakale igama elilodwa, ubuchopho kufanele buxhumanise amakhulukhulu emisipha yolimi, izindebe, izintambo zezwi, ne-diaphragm ngokulandelana okunembe kakhulu okuhlukaniseke kuma-millisecond. Lapho isiguli silahlekelwa ikhono elingokomzimba lokukhuluma, ubuchopho lusashisa le miyalo kagesi eyinkimbinkimbi. Ukuzama ukunquma leso sikhukhula sedatha ye-neural usebenzisa izibalo zendabuko kufana nokuzama ukufunda izincwadi eziyinkulungwane ngesikhathi esifanayo; isistimu iphahlazeka ngokushesha, idinga i-AI ethuthukisiwe ukuze ihlele idatha ibe amaphethini anengqondo.
A: Imfihlo yokugeleza kwengxoxo yemvelo yesistimu ilele ekubambezelekeni kwayo okuphansi kakhulu kwe-30-millisecond processing. Izinsiza zokuxhumana ezijwayelekile zidinga ukuthi umsebenzisi akhethe ngokucophelela izinhlamvu ngayinye ngayinye esebenzisa izilandeleli zamehlo, adale ukuthula okude, okukhathazayo okuphazamisa isigqi sokusebenzelana kwabantu okuvamile. Uhlaka lukaDkt. Stavisky lwezigaba ezimbili ze-AI lweqa lokhu kuthayipha kohlamvu nezinhlamvu ngokuphelele. Ngokusebenzisa imodeli yesendlalelo sokuqala ukuze kutholakale amayunithi omsindo ayisisekelo (amaphoneme) kanye nemodeli yesendlalelo sesibili (efana nemodeli yolimi olukhulu) ukuze abikezele amagama ahlosiwe ngokushesha, uhlelo luvala igebe phakathi komcabango nomsindo cishe ngokuphelele, okuvumela umsebenzisi ukuthi akhulume ngesivinini somcabango womuntu wemvelo.
A: Lesi senzakalo esiyingqopha-mlando simele ubufakazi obukhulu bomqondo wawo wonke umkhakha we-neuro-restoration. Ngokubonisa ukuthi isixhumi esibonakalayo esiqhutshwa yi-AI singahumusha izindlela ze-neural ezilimele zibe enkulumweni enembe kakhulu, ecacile, lolu cwaningo luvula indlela yomtholampilo eqondile yokusiza izigidi zabantu abathuliswe izimo ezihlukahlukene. Umqondo ofanayo wokukhipha amakhodi wezigaba ezimbili ekugcineni ungalungiswa kahle ukuze usize abasinde bestroke bazulazule nge-aphasia enzima, abantu abasha abaphethe i-cerebral palsy, noma iziguli ezilulama ekulimaleni okubuhlungu kobuchopho. Ifakazela ukuthi imephu yezinzwa yolimi ihlala injalo engqondweni; sidinga nje ukwakha amabhuloho edijithali alungile ukuze sivumele lawo mazwi angaphakathi abuyele emhlabeni.
Amanothi Omhleli:
- Lesi sihloko sihlelwe umhleli weNeuroscience News.
- Iphepha lejenali libuyekezwe ngokugcwele.
- Ingqikithi eyengeziwe yengezwe abasebenzi bethu.
Mayelana nalezi zindaba ze-AI ne-neurotech
Umbhali: Meagan Phelan
Umthombo: I-AAAS
Othintana naye: Meagan Phelan – AAAS
Isithombe: Isithombe sifakwe ku-Neuroscience News
