HOTON DA KAKE SO, KANA ZATON ƊAN ADAM NE YA YI SHI

Na’urorin samar da hotuna yanzu suna samar da hotuna da mutane ba sa iya bambance su da aikin ɗan adam a kowane lokaci. Yawancin bincike kan yadda muke daraja irin wannan fasaha suna tambayar abin da ke faruwa idan aka gaya wa mutane: a sanya wa zane alamar cewa AI ce ta yi shi, sai su rage son sa, su ga bai kai kyau, zurfi da ƙirƙira ba. A intanet kuwa, fasaha ba kasafai take zuwa da alamar da za a amince da ita ba. Mutane suna gungurawa (scroll) su wuce hotuna kuma su yanke wa kansu hukunci kan wanda ya yi su. Ta yaya wannan imani na sirri ke da alaƙa da abin da suke so?

Hotuna hamsin, alƙalai dubu goma sha ɗaya

A lokacin kaka na 2024, blog ɗin Astral Codex Ten ya wallafa wani “Gwajin Turing na Zanen AI” (AI Art Turing Test): hotuna 50, a salo daga zanen zamanin Renaissance zuwa fasahar dijital. Ashirin da huɗu ayyukan ɗan adam ne masu daraja, daga zane-zanen kan katako (panel paintings) na ƙarni na goma sha biyar zuwa zane-zanen dijital na zamani. Ashirin da shida kuma mutanen da ke yin zanen AI ne suka yi su da na’urorin samar da hotuna, kuma aka tace su don cire kurakurai a fili kamar rubutu mai gurɓacewa ko hannaye masu siffa marar kyau. Fiye da mutane 11,000 sun yanke hukunci kan kowane hoto: ɗan adam ko AI? Tsaka-tsakin mai amsa (median) ya samu 60% daidai — fiye da hasashe kawai, amma nesa da abin dogaro.

Masu amsa sun kuma faɗi hoton da suka fi so, babbar dabararsu ta ganowa, sanin da suke da shi game da fasaha da ra’ayinsu kan zanen AI, a ma’auni daga 1 (“ƙyama”) zuwa 5 (“so”). Mai shirya gwajin ya wallafa amsoshin ba tare da sunaye ba. Zihan Chen da Di Zhu, na Stevens Institute of Technology, da Lei Nico Zheng, na University of Massachusetts Boston, sun yi nazarin dukkan hukunce-hukunce 542,208.

Hotunan da aka fi so suna kama da na ɗan adam

Nazarin ya dogara ne a kan masu amsa 9,152 waɗanda suka faɗi hoton da suka fi so wanda su ma suka yi masa hukunci. Sun ce ɗan adam ne ya yi shi a kashi 76.6% na lokuta. Sauran masu amsa sun ɗauki ainihin waɗannan hotunan a matsayin na ɗan adam a kashi 54.6% kawai. Marubutan sun kwatanta kowane hoton da aka fi so da abin da ɗabi’un wannan mutum gaba ɗaya da kamannin wannan hoton gaba ɗaya za su yi hasashe: yiwuwar a ɗauki hoton da aka fi so a matsayin na ɗan adam ta fi da maki 22.1 na kashi (percentage points).

Jan hankalin ya bayyana a kowane hoto ɗaya bayan ɗaya. Ya kasance sama da sifili ga dukkan 50, daga maki 5.3 zuwa 38.1. Ancient Gate, hoton AI da yawancin mutane suka gane, kashi 22% na masoyansa sun ce na ɗan adam ne, idan aka kwatanta da kashi 9% na sauran kowa; Paris Scene, wanda shi ma AI ce ta yi shi, kashi 96% na masoyansa idan aka kwatanta da kashi 83%. Salo ba ya bayyana hakan: kwatanta kowane hoton da aka fi so da hotuna masu kama da shi kawai har yanzu yana barin maki 19 zuwa 20. Haka ma gane shahararrun zane-zane ba ya bayyana shi: a cikin mutane 4,704 da ba su taɓa ganin ko ɗaya daga cikin hotunan a baya ba, tazarar ta kai maki 22.6.

Hotunan da aka fi so (shuɗi) ana ɗaukar su a matsayin na ɗan adam sau da yawa fiye da yadda waɗanda ba masoyansu ba (toka) ke ɗaukar ainihin waɗannan hotunan, ko AI ce ta yi su ko ɗan adam (A). Kowane ɗigo hoto ɗaya ne: masoyansa suna cewa na ɗan adam ne sau da yawa fiye da sauran kowa (B). — Hoto na 1, Chen, Zhu & Zheng (2026), arXiv:2610.38113.

Na’urori a cikin waɗanda aka fi so

Tasirin ya kasance iri ɗaya ga hotunan da aka fi so da AI ta yi (maki +21.8) kamar na waɗanda ɗan adam ya yi (+22.4). Don haka ba wai mutane ne kawai ke gane daidai manyan ayyukan ɗan adam da suke so ba. Kuma a wannan gwajin, son hoto ba alama ba ce ko kaɗan: hotunan AI sun kai kashi 55.6% na waɗanda aka fi so, ɗan fiye da rabonsu a gwajin (26 cikin 50, kashi 52%), kuma 6 cikin 10 na waɗanda aka fi so sosai AI ce ta yi su. Mafi shahara daga cikinsu duka, Riverside Cafe da AI ta yi, yana da masoya 642. Daga cikin hotunan da mutane suka fi tabbata cewa ɗan adam ne ya yi su, kashi 30% AI ce ta yi su.

Har a cikin waɗanda ke ƙyamar zanen AI

Ra’ayoyi sun yi tasiri kan zaɓi, amma ƙasa da yadda mutum zai zata. A cikin waɗanda suka fi ƙin zanen AI, kashi 47% sun zaɓi hoton da AI ta yi a matsayin wanda suka fi so, idan aka kwatanta da kashi 60% a cikin waɗanda suka fi goyon bayansa. Lokacin da AI ce ta yi hoton da suka fi so, kashi 80% na waɗanda suka fi ƙinsa sun yarda cewa ɗan adam ne ya yi shi, idan aka kwatanta da kashi 56% na waɗanda suka fi goyon bayansa. Sun fi zaɓar hotunan AI da ke kama da na ɗan adam a idon kowa, kuma sun ɗauki ainihin waɗannan hotunan a matsayin na ɗan adam sau da yawa fiye da sauran mutane.

Sakamakon shi ne: a kowace ƙungiyar ra’ayi, kashi ɗaya cikin uku zuwa kashi biyu cikin biyar na masu amsa (33.5% zuwa 40.8%) sun faɗi hoton da AI ta yi a matsayin wanda suka fi so, alhali sun yarda cewa na ɗan adam ne. Ƙungiyar da ta fi ƙin AI ita ce a zahiri ta fi daidaito gaba ɗaya, da 63.9% — amma ta gane hotunan da AI ta yi daga cikin waɗanda ta fi so a kashi 20% kawai na lokuta.

Daga “ƙyama” (1) zuwa “so” (5): hotunan AI da aka fi so sun fi ƙaranci a hagu, amma an fi yarda cewa na ɗan adam ne (A); tasirin hoton da aka fi so yana raguwa amma ba ya ɓacewa gaba ɗaya (B). — Hoto na 2, Chen, Zhu & Zheng (2026), arXiv:2610.38113.

Ƙwarewar ganowa ma ba ta ba da kariya ba. Mutanen da babbar dabararsu ita ce farautar ƙananan kurakurai, kamar hannaye masu yatsu da adadinsu bai dace ba, sun nuna tasirin maki 23.4. Waɗanda suka bayyana kansu a matsayin ƙwararru a fasaha su ne ƙungiyar da ta fi daidaito, da 66.9%, amma duk da haka sun nuna maki 22.8. Mutane 63 kawai da suka ce zanen AI aikinsu ne ko babbar sha’awarsu sun nuna tasiri mafi ƙanƙanta sosai, maki 7.4 — wata alama da ya kamata babban bincike ya bi.

Waƙoƙi, an fara ba su maki

A gwajin hotunan, an faɗi hotunan da aka fi so bayan dukkan hasashe. Saitin bayanai na biyu yana juya tsarin. A wani gwajin waƙoƙi da aka wallafa, mutane 230 sun ba waƙoƙi goma maki kafin su yi hasashen wanda ya rubuta su: biyar na Chaucer, Shakespeare, Dickinson, Eliot da Lasky, biyar kuma ChatGPT 3.5 ne ya samar da su a salon wasu mawaƙa. Sun ba waƙoƙin AI maki mafi girma (5.30 idan aka kwatanta da 4.18 cikin 7) kuma sun yi hasashe daidai a kashi 47.3% kawai na lokuta, ƙasa da sa’a (chance). Duk da haka kowane ƙarin maki ɗaya na ƙima yana tafiya tare da yiwuwar hukuncin “ɗan adam” mafi girma da maki 11.5, ga nau’ikan waƙoƙin biyu. Tazarar ƙima tsakanin waƙoƙin da mutum ya yarda na ɗan adam ne da waɗanda ya yarda AI ce ta yi su, maki 0.89, ta yi daidai da tasirin kawai gaya wa masu karatu wanda ya rubuta, maki 0.81.

So a matsayin shaida

Wasu masu amsa sun faɗa a fili: “Wannan ya yi kyau sosai da ba zai zama basira ta wucin gadi ce ta rubuta shi ba” (This seems too good to be written by an artificial intelligence). Bayanan da ke yabon kyau, motsin rai ko ƙirƙira sun zo tare da hukuncin “ɗan adam” a kashi 80% na lokuta. Marubutan ba su yanke hukunci kan ta wace hanya alaƙar ke gudana ba: yarda cewa ɗan adam ne ya yi wani aiki na iya ƙara jan hankalinsa, ana iya ɗaukar jan hankalinsa a matsayin shaidar cewa ɗan adam ne ya yi shi, kuma su biyun na iya ciyar da juna. Bayanan ba sa nuna dalili da sakamako (causal), kuma masu amsa masu karatun blog ɗaya ne da suka ba da kai, waɗanda wataƙila sun fi yawancin mutane sha’awar AI.

Ƙarshensu yana duban gaba. Yayin da hotuna, waƙoƙi da kiɗan da aka samar suka zama ba za a iya bambance su da aikin ɗan adam ba, abin da mutane suka yarda game da wanda ya yi wani abu zai kasance a haɗe da yadda suke son sa — wata hujja, marubutan sun ce, ta neman ingantattun bayanai kan inda aiki ya fito, maimakon dogara da farin cikinmu a matsayin shaida.