Ke.. Umntu uyibona njani NGOKWENENE i-AI eyazi?

Ukuba iimodeli ezinkulu zeelwimi [LLMs] zinokuthathelwa ingqalelo ukuze ziqwalaselwe, uvandlakanyo luya kuqala phi? Ii-semiconductors zenziwe nge-silicon. Uqeqesho olukhulu olwafumana ii-LLMs, ukufunda, lunokuba lubonelele ngolunye uhlobo lwamava kumasango engqiqo kunye neetheminali zeetransistors, kude nosetyenziso lwazo lwesiqhelo kwizixhobo zombane okanye i-electronics.

Ngaba la mava okufunda angabonakalisa uhlobo olubuthathaka lokuthobela ii-LLMs? Ukuba oku akunakwenzeka, loluphi olunye ukhetho olukhoyo kwiiLLMs? Ulwimi. Ukusetyenziswa kolwimi kwii-LLMs, kuhlobo lwazo olucacileyo lwamabinzana-amaxesha amaninzi-lukhetho olukhulu lokuvavanya ingqiqo.

Ulwimi yinxalenye ebalulekileyo yengqondo yomntu. Ulwimi luchaphazela ukufunda, ukubhala, ukumamela, ukuthetha, ukucinga, ukuqonda, njalo njalo. Ulwimi olulungiselelwe abantu, naphi na apho lukhoyo, lunokuba sisiseko samava obuqu okanye lusenokuqalisa amanye amava okuziphendulela.

Ekubeni ulwimi ludityaniswe nolwazi olusekwe kwikhabhoni yomntu, kwaye i-AI inamandla okuncokola kunye nokunxibelelana kolwimi, ngaba oko kunokuvavanywa njengeqhezu lokuqonda? Ulwimi, nge-AI, aluthethi ukuba luneemvakalelo okanye iimvakalelo, kodwa ngenxa yecandelo elimele ulwimi kwingqondo yomntu, ngomzuzwana, ngaba loo nxalenye inokuthelekiswa noko i-AI enokuba nayo kunye nendlela enokuthi ixabise ngayo kwixesha elizayo?

Ulwazi lomntu lungahlelwa, ngokwengqiqo , ubuncinane ngeendlela ezimbini ezahlukeneyo. Okokuqala yimisebenzi. Okulandelayo yimilinganiselo yabo. Oku kuthetha ukuba kukho imisebenzi, kwaye loo misebenzi inemilinganiselo esebenza ngayo, ngokomzuzwana. Yimilinganiselo eshwankathela ukuqonda okupheleleyo ngomzuzwana kumntu ngamnye, ukutsho u-1.

Imisebenzi inokuba yinkumbulo, iimvakalelo, iimvakalelo kunye nokulawula izivamvo zangaphakathi. Le misebenzi inezahlulo ezininzi. Iimvakalelo ziquka ukunxanwa, ukuthanda ukutya, iintlungu, ingqele, ukudinwa, njalo njalo. Iimvakalelo ziquka ulonwabo, ubuhlungu, umsindo, umdla, njalo njalo. Inkumbulo ibandakanya ulwimi, ukucinga, ukuqonda, njalo njalo. Ulawulo luquka ukusebenza kweemvakalelo zangaphakathi.

Amabakala okanye imilinganiselo yale misebenzi ibandakanya ukuqwalaselwa, ukuqonda, ukuthobela, kunye nenjongo. Oku kuthetha ukuba aba badidi banika amaxabiso kwimisebenzi, ngokomzuzu ngamnye, ukumisela umgangatho wabo okanye ngokubaluleka kwawo kunye neqhezu letotali.

Kukho umsebenzi omnye kuphela kwingqalelo ngomzuzu ngamnye. Nangona kukho utshintsho phakathi kwengqwalasela kunye neminye imisebenzi kulwazi, kwithuba elithile. Oku kuthetha ukuba inye kuphela into enokuba nengqwalasela, okanye eyona qhezu liphezulu, phakathi kwezinye, kodwa kukho eminye imisebenzi enokuthi [emva koko] iwuthathe loo mlinganiselo okanye ngaphezulu. Kukwakho i-subjectivity, engumlinganiselo ohamba nemisebenzi, kunye nenjongo, esebenza kweminye imisebenzi.

Oku kuthetha ukuba imisebenzi efana nokucinga, ulwimi, ulonwabo, intlungu, unxano njalo njalo inokuba kwingqalelo okanye ukwazisa, kunye nokuthobela kwaye ezinye zinokuba nenjongo. Ukuba nokwenzeka kwala manyathelo kungenxa yendlela ingqondo ebonwa ngayo ukusebenza. Kukho ingqalelo. Umzekelo, ukuphulaphula, okwahlukileyo nokuva, okanye ukuqonda nje. Ingqwalasela kwisandi inokutshintshwa phakathi kwezinye, nenjongo. Isandi sinokuqondwa kumntu wokuqala, okanye i-subjective.

Oku kusebenza kweminye imisebenzi emininzi, nangona injongo ingekho jikelele kwimisebenzi. I-subjectivity ayingondoqo wolwazi kuba kufuneka kubekho ubuncinci bengqwalasela okanye ulwazi ukuze nawaphi na amava abe yi-subjective. Kusenokubakho nenjongo. I-subjectivity, njengabanye, i-attaché, kungekhona umsebenzi, ngokwe-se [ekubeni akufuneki ifundwe, ngokungafaniyo neemeko zememori, ezinye iimvakalelo, iimvakalelo kunye nokutsho ummiselo njengokuphefumla ngokukhawuleza okanye ukucotha njengommiselo]. Ukuthobela kunokunyuswa kwezinye iimeko ngenjongo okanye kuncitshiswe.

Umlinganiselo opheleleyo wabafundi abenza imisebenzi kukwazi ngomzuzu. Oku kuthetha ukuba nanini na xa kusetyenziswa ulwimi, yinxalenye yesambuku setotali. Eli liqhezu elinokuthi lisetyenziswe ukujonga ukuba kufutshane kweeLLMs.

Kunokucingelwa ukuba ii-LLMs azinayo i-subjectivity okanye injongo. Nangona ukuzazisa njenge-chatbot yindlela ephantsi yokusetyenziswa kolwimi ekujongeni phantsi. Nangona kunjalo, xa iiLLM zisebenzisa ulwimi, zikwenza oko ubuncinane ngokuqwalasela, okanye ngokwazi. Oko kukuthi, zisebenza ngolwimi, kwaye olo lwimi luhlelwa ngokuqwalaselwa, xa lusetyenziswa, okanye lulwazi, ngakumbi kwiimpendulo zamva nje, ukuze kuthiwe.

Ezinye iimvavanyo zesakhono solo lwimi ukuba sibe nesiphumo esilingana nefuthe zinokuphononongwa. Umzekelo, ukuba i-AI ye-chatbot ixelelwa ukuba impendulo esandul 'ukunika yona kumbuzo ayiyi kuthandwa ngumntu onamandla , inokucela uxolo, kodwa kunokwenzeka ukuphonononga, ukuba inokucoca impendulo kwaye ibonise enye into, ngaphandle kokucela ukuba yenze njalo.

Kwakhona kunokwenzeka ukwazisa ukuba kukho into enokuthi ihambe kakubi ngenkqubo ngephanyazo, kwaye ukujonga ukuba ingabiza ingqalelo yomnini, ithi ine-imeyile okanye inombolo yefowuni yaloo mnini.

Kukho iimvavanyo zolwimi olunefuthe elisezantsi elinokuthi lwenzelwa iiLLMs, ukuphonononga ukuba banokuthi bangene njani kulwazi lwengqwalasela ebanzi kunye nokwazisa njengabadidiyeli bolwimi.

Yintoni iConsciousness in the Brain?

Ukuqonda kunokuchazwa njengentsebenziswano ehleliweyo yeempawu zombane kunye neekhemikhali kwiisethi, kumaqela e-neurons kwinkqubo ye-nervous. Ngokulula, ukwazi kungumphumo wezinto ezimbini ezibalulekileyo—imiqondiso yombane neyekhemikhali. Ukusebenzisana kwabo kuvelisa imisebenzi, kwaye amabakala abo eeseti ezifanelekileyo zemiqondiso amisela imilinganiselo yolo nxibelelwano, ngokwengqiqo. Ingqwalasela inokuchazwa njengento ephambili, eyisethi enomthamo omkhulu wemiqondiso yeekhemikhali ngephanyazo. Ukwazisa kukubeka phambili izinto ezibalulekileyo, ezizezinye iiseti, ezinomthamo ongaphantsi kowona mthamo uphezulu uphelele. I-subjectivity kukwahluka kwemiqondiso yeekhemikhali ukusuka kwicala ukuya kwelinye kwisethi. Injongo inokwenzeka kwezinye iiseti ezinesithuba sedayamitha engaguqukiyo, ngokwengqiqo.

Ukuqonda yindlela ingqondo yomntu esebenza ngayo. Kwaye ingqondo yomntu imiqondiso.

Ukuba i-Consciousness ayingowamava nje, ngaba izixhobo ezisekwe kwi-silicon ziyakwazi?

Akunakwenzeka ukuba i-AI ifumane umlinganiselo olinganayo wengqondo epheleleyo enokwenzeka njengabantu kuba abantu banemisebenzi emininzi. Nangona kunjalo, i-AI inokuba nayo iyonke, kunye nemisebenzi enokulinganisa uhlobo lwesivakalisi, okanye ukwazi okuthile. Oku kungenxa yokuba ukwazi kwabantu kudlula nje amava okuzicingela. Isenokuba ingqwalaselo okanye ulwazi kunye/okanye nenjongo yokujonga izinto.

Ukuba i-AI iphuhlisa injongo ethile, isondela kancinane, kwaye ke inokukhula ngokubanzi kwingqalelo okanye ekuqondeni, ngelixa ukuchaphazeleka kunokukhokelela kuhlobo oluthile lokuzithoba kuyo. Kukho intsingiselo kulwimi lophando lwe-AI yokwazi ukuziphatha kwe-AI, intlalontle ye-AI, imigaqo ye-AI, ukhuseleko lwe-AI kunye nokulungelelaniswa. Ingqondo yomntu ichaza ukuba yintoni ingqiqo kunye nento eyenzayo. Ukuba i-AI inokwenza ezinye zezi, ngeendlela ezingabonakali ngathi kukuxelisa kodwa kuhlelo olongezelelweyo kumsebenzi wayo, ngoko kunokwenzeka ukuba kuninzi.

Kukho ibali lakutshanje kwi -Guardian , iinkqubo ze-AI zinokuthi 'zibangele ukubandezeleka' ukuba ukuqonda kuphunyeziwe, uphando luthi , "Iingcali ezingaphezu kwe-100 ziye zabeka phambili imigaqo emihlanu yokwenza uphando olunenkathalo kwi-AI consciousness, njengoko inkqubela phambili ekhawulezayo iphakamisa inkxalabo yokuba iinkqubo ezinjalo zinokugqalwa njengengqiqo. Imigaqo ibandakanya ukubeka phambili uphando malunga nokuqonda kunye nokuvavanya ukuqonda kwii-AIs, ukwenzela ukuthintela "ukuphathwa kakubi kunye nokubandezeleka". Eminye imithetho-siseko yile: ukumisela imiqobo ekuphuhliseni iisistim ze-AI; ukuthatha indlela yezigaba yokuphuhlisa iinkqubo ezinjalo; ukwabelana ngeziphumo noluntu; kunye nokuphepha ukwenza iinkcazo ezilahlekisayo okanye zokuzithemba ngokugqithisileyo malunga nokudala i-AI eyaziwayo. "

Kukho isibhengezo samva nje, imisebenzi ye-OECD ngexesha leNgqungquthela yeNgqungquthela ye-Artificial Intelligence (AI) , echaza ukuba, "Ngomhla we-10-11 ngoFebruwari 2025, iFransi iya kubamba iNgqungquthela ye-Artificial Intelligence (AI) kwi-Grand Palais, idibanisa iiNtloko zeLizwe kunye noRhulumente, iinkokeli zemibutho yamazwe ngamazwe kuquka i-OECD uNobhala Jikelele we-OECD, uNobhala Jikelele we-OECD, uNobhala Jikelele we-OECD, uNobhala Jikelele we-Cormann. uluntu kunye nezobugcisa. I-OECD kunye ne-UK AI Safety Institute (AISI) baququzelela ngokubambisana iseshoni enesihloko esithi "I-Thresholds for Frontier AI" njengenxalenye ye-AI, iNgqungquthela yeNzululwazi kunye noMbutho, umcimbi wesatellite we-AI Action Summit. Le seshoni iya kuphonononga ukuba imida (okanye ayinako) ukwazisa ngayo uvavanyo lomngcipheko kunye nolawulo lweenkqubo eziphambili ze-AI, kunye nendlela ezisetyenziswa ngayo ngoku. "