Haɗin Gwiwar Ba da Labari tare da ƴan Wasa na Mutane da Masu Ba da Labari na AI: Nazarin Rahoton Taron

1. Gabatarwa & Bayyani

Wannan rahoton taron ya ƙididdige wani gwaji na farko a cikin haɗin gwiwar ƙirƙira tsakanin mutum da AI, inda aka sanya babban samfurin harshe (GPT-3) a matsayin mai ba da labari tare a cikin yanayin wasan kwaikwayo na raye-raya na kai tsaye wanda ba a rubuta shi ba. Babban manufar ita ce bincika ko AI zai iya bin ci gaban labari da kuma yanayin halaye a ainihin lokaci, yana ba da jagorar labari wanda ƴan wasan mutane za su iya fassara su kuma su yi. Bayan ya wuce hulɗar ta zamani ko ta wasa, wannan aikin ya kawo AI cikin yanayin zahiri na zamantakewa na dandamali, yana gwada iyawarsa don haɗin gwiwa mai ma'ana a cikin tsarin ƙirƙira mai matuƙar mahimmanci da kuma na bazata.

Aikin ya sanya wasan kwaikwayo na raye-raya a matsayin "gurin gwaji" na musamman don kimanta hankalin zamantakewa da na labari na AI, yana mai da hankali kan iyawar samfurin na "ba da hujja" da kuma daidaitawa da abubuwan labari masu tasowa—wata babbar ka'ida ta aikin raye-raya.

2. Hanyoyin Bincike & Tsarin Tsarin

An tsara tsarin don sauƙaƙe tattaunawa tsakanin ƴan wasan raye-raya na mutane da mai ba da labari na AI. Matsayin AI ba shine yin wasa a matsayin hali a kan dandamali ba, amma ya zama jagoran labari a wajen dandamali, yana ba da saitin fage, juyoyi, da kuma ƙarshe.

2.1. Mai Ba da Labari na AI: Haɗin GPT-3

Ƙungiyar ta yi amfani da GPT-3 na OpenAI, samfurin harshe na transformer wanda aka sani da ƙarfin iyawarsa na koyo kaɗan (few-shot) da kuma ba tare da misali (zero-shot). An ƙarfafa samfurin don samar da abun ciki na labari bisa ga yanayin ci gaba na wasan kwaikwayo. Mahimmin abin da ke cikin aikinsa shine kiyaye daidaiton labari a kan dogon labari mai tasowa.

2.2. Tsarin Aiki & Ƙuntatawa

An gabatar da sabbin ƙuntatawa don karkatar da GPT-3 daga gajerun amsoshi na tattaunawa zuwa dogon bayani na labari wanda ya dace da fage na wasan kwaikwayo. Wannan mai yiwuwa ya ƙunshi dabarun ƙirƙira umarni waɗanda ke ƙayyade tsawon fitarwa, sautin labari (misali, "mai siffantawa," "mai ban mamaki"), da kuma nassoshi kai tsaye zuwa abubuwan da suka gabata na labari don tabbatar da ci gaba.

3. Tsarin Gwaji & Ayyukan Kai Tsaye

Aikin ya ci gaba ta hanyar tsari na ci gaba da gwaji, wanda ya ƙare a cikin ayyukan jama'a.

3.1. Lokacin Atisaye tare da Ƙwararrun ƴan Raya-raya

An fara gwada tsarin AI a cikin atisaye tare da ƙungiyar ƙwararrun ƴan raye-raya. Wannan lokacin yana da mahimmanci don maimaita ƙuntatawar samfurin, fahimtar yadda ƴan wasa suka fassara labarin da AI ya samar, da kuma inganta aikin haɗin gwiwar mutum da AI. Ya zama wurin gwaji don daidaita gudunmawar AI don zama mai ƙarfafa ƙirƙira amma mai yiwuwa ga wasan kwaikwayo na kai tsaye.

3.2. Ayyukan Kai Tsaye na Jama'a a Bikin Wasan Kwaikwayo

An gwada tsarin a cikin ayyukan kai tsaye guda biyu don masu sauraron jama'a a matsayin wani ɓangare na bikin wasan kwaikwayo na Turai. Wannan ya ba da ingantattun yanayi masu matsin lamba don kimanta ƙarfin tsarin da kuma karɓuwar masu sauraro na ba da labari ta hanyar AI.

4. Sakamako & Kimantawa

An gudanar da kimantawa ta hanyar binciken bayan wasan kwaikwayo na membobin masu sauraro da ƴan wasa, yana ba da hangen nesa biyu akan ingancin AI.

Mahimman Ma'auni na Ra'ayi

Zaɓin Masu Sauraro: Kyakkyawar amsa ga ba da labari na AI; sun nuna fifita AI a matsayin mai ba da labari fiye da AI a matsayin hali a kan dandamali.
Karɓuwar ƴan Wasa: Sun amsa da kyau; sun nuna sha'awar sabbin hanyoyin labari masu ma'ana da ƙirƙira waɗanda AI ya gabatar.
Tabbatar da Tsarin: Sakamakon ya goyi bayan wasan kwaikwayo na raye-raya a matsayin wurin gwaji mai amfani don bincika haɗin gwiwar mutum da AI a cikin yanayin zamantakewa.

4.1. Ra'ayoyin Binciken Masu Sauraro

Masu sauraro sun amsa da kyau ga gogewar. Zaɓin ba da labari na AI fiye da shigar da hali na AI yana nuna cewa masu sauraro sun fi karɓar AI a matsayin jagora, matsayin labari na sama (kamar marubucin wasa ko darekta) fiye da zama wakili na zamantakewa kai tsaye, wanda har yanzu zai iya faɗowa cikin "kwarin ban mamaki" na hulɗa.

4.2. Ra'ayoyin ƴan Wasa & Tasirin Ƙirƙira

Ƴan wasa sun ba da rahoton cewa AI ya gabatar da juyoyin labari da ba a zata ba masu ban sha'awa, suna tura su daga yankin kwanciyar hankalinsu ta hanya mai fa'ida. Wannan ya yi daidai da ka'idar raye-raya ta "ba da hujja," inda ƴan wasa dole ne su daidaita da sabbin tayin cikin ƙirƙira. AI ya yi nasara a matsayin tushen irin waɗannan tayin, yana haɓaka maimakon hana ci gaban ƙirƙira.

5. Cikakkun Bayanai na Fasaha & Ƙuntatawar Samfurin AI

Babban ƙalubalen fasaha shine daidaita samfurin harshe na gabaɗaya (GPT-3) zuwa takamaiman yanki na samar da labari mai tsayi, mai daidaituwa. Rahoton ya ambaci "sabbin ƙuntatawa" don samar da dogon rubutu na labari. Wannan mai yiwuwa ya ƙunshi haɗuwa da:

Ƙirƙiran Umarni: Ƙirƙiran umarnin tsarin waɗanda suka ayyana matsayin AI (misali, "Kai ne mai ba da labari na wasan kwaikwayo..."), sun ƙayyade tsarin fitarwa, kuma sun haɗa da misalan salon labari da ake so.
Gudanar da Mahallin: Ciyar da samfurin taƙaitaccen tarihin labarin da ke gudana don kiyaye daidaituwa. Hanyar kulawa a cikin masu canzawa kamar GPT-3 ana iya siffanta su kamar $\text{Kulawa}(Q, K, V) = \text{softmax}(\frac{QK^T}{\sqrt{d_k}})V$, inda $Q$, $K$, $V$ su ne tambayoyi, makullai, da ƙimomi waɗanda aka samo daga jerin shigarwa. Yanke mahallin yana da mahimmanci don ci gaba da kasancewa cikin iyakokin alama yayin da ake adana bayanan mahimmanci na makirci.
Ƙuntataccen Fassarawa: Mai yiwuwa yin amfani da dabarun don karkatar da samarwa zuwa wasu batutuwa, guje wa maimaitawa, ko tilasta mafi ƙarancin tsawon fitarwa.

Zanen Tsarin Aiki na Hasashe: 1. Ƴan wasan mutane sun kammala wani fage. 2. Mai sauƙaƙawa na mutum (ko tsarin atomatik) ya taƙaita mahimman abubuwan makirci/halaye. 3. An tsara wannan taƙaitaccen bayani zuwa umarni don GPT-3. 4. GPT-3 ya samar da bugun labari na gaba (misali, "Ba zato ba tsammani, mai binciken ya tuna da wasiƙar da aka ɓoye a cikin littafin..."). 5. An ba da labarin ga ƴan wasa (ta allon ko kunnen kunne). 6. Ƴan wasa sun ba da hujja kuma suka yi sabon tayin na labari.

6. Tsarin Nazari & Misalin Lamari

Tsari: Matrix na "Daidaiton Labari & Walƙiya na Ƙirƙira"
Wannan tsari yana kimanta haɗin gwiwar AI a cikin ba da labari tare da ginshiƙai biyu:

Daidaiton Labari (X-axis): Iyawar AI na kiyaye daidaiton makirci na ma'ana, ƙwaƙƙwaran hali, da alaƙar dalili da sakamako.
Walƙiya na Ƙirƙira (Y-axis): Iyawar AI na gabatar da sababbin ra'ayoyi, waɗanda ba a zata ba, masu ban sha'awa waɗanda ke tura labarin zuwa sabbin hanyoyi masu ban sha'awa.

Misalin Lamari: A cikin atisaye, ƴan wasan mutane sun kafa wani fage game da masu dafa abinci guda biyu suna jayayya kan girke-girke. Shigar da mai ba da labari na AI shine: "Ba su sani ba, sirrin sinadarin da suke faɗa a kansa a zahiri kayan yaji ne da aka sace daga ɗakin dafa abinci na sarauta shekaru da suka wuce. Wani siffa mai duhu yana kallon su daga lungu." Wannan motsi yana da maki mai girma akan Walƙiya na Ƙirƙira (gabatar da asiri, tarihin baya, sabon hali) yayin da yake kiyaye Daidaiton Labari ta hanyar haɗa rikici zuwa babban makirci na ma'ana. Sai ƴan wasan mutane suka ba da hujja ga wannan ta hanyar ɗaya daga cikin masu dafa abinci ya kalli tagar cikin tsoro, nan da nan ya ɗauki halin tsoro, don haka ya haɗa tayin AI cikin sauƙi.

7. Bita Mai Ma'ana na Manazarta

Babban Fahimta: Wannan aikin ba kawai game da AI yin raye-raya ba ne; yana da kyakkyawan gwajin damuwa don hankalin labari a cikin LLMs, ta yin amfani da tanderun kai tsaye na wasan kwaikwayo na kai tsaye wanda ba ya gafartawa. Haɓakar gaske ita ce binciken cewa AI-a matsayin mai ba da labari yana aiki da kyau fiye da AI-a matsayin ɗan wasa. Wannan yana bayyana babban fahimta game da ƙarfin AI na yanzu: yana da ƙarfin samar da ra'ayi da kuma ginin tsarin, amma yana kuskure a cikin ƙwaƙƙwaran, jiki, raye-rayen zamantakewa na hulɗa kai tsaye. Zaɓin masu sauraro ya tabbatar da cewa muna amincewa da AI a matsayin "fatalwa a cikin na'ura" yana ba da wahayi fiye da zama mutum-mutumi mai raba dandamali.

Tsarin Ma'ana: Ma'anar bincike tana da inganci: 1) Gano babban injin raye-raya (tayi & ba da hujja) a matsayin gwaji mai kyau don daidaitawar AI. 2) Sanya AI a matsayin da ya fi dacewa da iyawarsa na yanzu (mai ba da labari, ba ɗan wasa ba). 3) Yin amfani da ƙwararrun ƴan wasa a matsayin masu tacewa da masu fassara don fitarwar AI. 4) Tabbatarwa a cikin mafi ingantaccen saiti mai yiwuwa—masu sauraro kai tsaye. Wannan yayi daidai da falsafar ƙira mai maimaitawa da ake gani a cikin nasarar binciken hulɗar mutum da kwamfuta, kamar zagayowar ƙira mai da hankali kan mai amfani waɗanda cibiyoyi kamar MIT Media Lab ke ba da shawarwari.

Ƙarfi & Kurakurai: Ƙarfi: Ingantaccen ingancin muhalli. Amfani da bayanan wasan kwaikwayo na kai tsaye zarin zinariya ne idan aka kwatanta da binciken dakin gwaje-gwaje. Mayar da hankali kan haɗin gwiwa fiye da kwaikwayi (gwajin Turing) hanya ce ta balagagge kuma mai fa'ida ga binciken AI. Kurakurai: Rahoton ba shi da cikakkun bayanai na fasaha—menene ainihin "sabbin ƙuntatawa"? Ta yaya aka auna daidaiton labari da ƙima? Ba a ƙididdige hanyar bincike da girman samfurin ba, wanda ya bar sakamako mai kyau a matsayin labari. Haka kuma ya wuce gona da iri ga gazawar da ba makawa: me ya faru lokacin da AI ya ba da tayi marar ma'ana ko kuma mai sabani? Sau nawa mai sauƙaƙawa na mutum dole ne ya shiga tsakani?

Fahimta Mai Aiki: Ga masu binciken AI: Ƙara ƙarfi kan tsarin mai ba da labari/edita/darekta don ƙirƙira AI. Zuba jari a cikin samfuran mahalli masu tsayi da ingantattun gine-ginen ƙwaƙwalwar labari. Ga masu fasaha da furodusoshi: Wannan kayan aiki ne mai yiwuwa, na gaba kusa. Fara gwaji da AI a matsayin mai tayar da hankali a cikin ɗakunan marubuta da tarurrukan atisaye yanzu. Kayan aikin ba maye gurbin ba ne amma mai haɓakawa. Ga masu ilimin ɗabi'a: Fara tsara jagororin don gudunmawar AI a cikin fasahar haɗin gwiwa—batutuwan marubuci, son zuciya a cikin samar da labari (shin AI ya karkata zuwa wasu dabi'u?), da kuma tasirin tunani akan ƴan wasa suna ɗaukar umarni daga na'ura suna buƙatar tattaunawa mai himma.

8. Ayyukan Gaba & Hanyoyin Bincike

Ƙarfafa Ƙwarewar Samfuri: Daidaita LLMs akan manyan tarin wasannin kwaikwayo, rubutun allo, da ka'idar labari don haɓaka samfuran "ban mamaki" na takamaiman yanki, kamar yadda aka daidaita Codex don lamba.
Haɗin Nau'i-nau'i: Haɗa alamun gani daga wasan kwaikwayo (ta hanyar isar da kyamara) ko alamun rayuwa na ɗan wasa don barin mai ba da labari na AI ya amsa ga sautin motsin rai ko jiki na fage.
Dandamalin Ba da Labari Mai Hulɗa: Girman wannan ra'ayi zuwa gogewa ta kan layi mai hulɗa, abubuwan ba da labari na haɗin gwiwa da ake watsawa kai tsaye, ko kayan aikin ƙirƙira labari na AI da aka keɓance don marubuta.
Amfani da Magani da Ilimi: Aiwatar da tsarin zuwa maganin wasan kwaikwayo ko saitunan ilimi, inda mai ba da labari na AI zai iya jagorantar mahalarta ta hanyar tsarin zamantakewa ko na tarihi.
Bincike akan Ƙwaƙwalwar Labari na Dogon Lokaci: Haɓaka tsarin AI waɗanda za su iya sarrafa hadadden makircin labari a cikin tsayin lokaci mai tsayi, babban ƙalubale da wannan aikin ya nuna kuma yana da mahimmanci ga ci-gaban aikace-aikace kamar wasannin bidiyo masu hulɗa ko ƙirƙira abun ciki na jerin.

9. Nassoshi

Brown, T.B., da sauransu. (2020). Samfuran Harshe Ƴan Ƙwararrun Malamai ne. Ci gaba a cikin Tsarin Sarrafa Bayanai na Neural, 33.
Mathewson, K., & Mirowski, P. (2017). Wasan Barkwanci na Raya-raya a matsayin Gwajin Turing. Proceedings of the AISB Symposium on AI and Society.
Mathewson, K., & Mirowski, P. (2018). Ƙirƙirar Labari na Lissafi na Raya-raya a Duniyar Gaske. Proceedings of the International Conference on Computational Creativity.
Riedl, M. O., & Stern, A. (2006). Wakilai Masu Imani da Daidaitawar Labari Mai Hankali don Ba da Labari Mai Hulɗa. International Conference on Technologies for Interactive Digital Storytelling and Entertainment.
Eger, M., & Mathewson, K. (2018). dAIrector: Samar da Bugun Labari ta Atomatik ta hanyar Haɗa Ilimi. Workshop on Intelligent Narrative Technologies.
Johnstone, K. (1979). Impro: Raya-raya da Gidan wasan kwaikwayo. Routledge.
Vaswani, A., da sauransu. (2017). Kulawa Duk Abinda Kake Bukata. Ci gaba a cikin Tsarin Sarrafa Bayanai na Neural, 30. (Don Transformer/Hanyar Kulawa mahallin).
MIT Media Lab. (n.d.). Bincike kan Yanayin ɗan Adam & Hulɗar Mutum da Kwamfuta. An samo daga media.mit.edu. (An ambata don falsafar ƙira mai maimaitawa).