MAKIN DAMUWA DA AI KE BAYARWA YA DOGARA DA AI ƊIN KANSA

Baƙin ciki mai tsanani (depression) ba shi da gwajin jini na tantancewa. Manyan samfuran harshe (large language models) suna alƙawarin tantancewa ba gajiya, mai rahusa kuma mai maimaituwa daga tattaunawar asibiti, kuma kimantawar farko sun ƙarfafa gwiwa. Ilimin taɓin hankali (psychiatry) ya taɓa fuskantar irin wannan tambaya a baya: a shekarar 1971, wani bincike ya gano cewa likitocin taɓin hankali na Amurka da na Birtaniya sun saɓa bayan sun kalli hirarraki iri ɗaya da aka naɗa a bidiyo, kuma fannin ya amsa da ƙa’idoji bayyanannu da hirarraki masu tsari.

Amma samfurin harshe yana zama “mai tantancewa” ne kawai ta hanyar jerin zaɓuɓɓuka. Wani ne ke zaɓar samfurin, ya tsara buƙatar — a matsayin takardar tambayoyi, da muryar likitan taɓin hankali, ko a matsayin mara lafiyar yana amsawa game da kansa — ya zaɓi lambobi ko haruffa don amsoshin, ya yanke shawara ko samfurin zai karanta kalaman mara lafiyar kaɗai ko dukkan tattaunawar, kuma ya mayar da abin da ya fitar zuwa maki. Kimantawa ba safai suke nuna waɗannan madadin ba.

Hanyoyi 880 na gina mai tantancewa ɗaya

Baihan Lin, na Icahn School of Medicine a Mount Sinai a New York, ya haɗa buɗaɗɗun samfura 11 (daga biliyan 1 zuwa biliyan 21 na sigogi) da hanyoyin furuci biyar, hanyoyi biyu na kallon tattaunawar, hanyoyi biyu na karanta amsar da hanyoyi huɗu na gina makin. Wannan ya ba da masu tantancewa 880, kowanne an yi amfani da shi kan hirarraki 189 da aka naɗa iri ɗaya, waɗanda wani mai yin hira na kwamfuta mai motsi ya gudanar. Kafin kowace hira, mahalarta sun cike PHQ-8, takardar tambayoyi kan baƙin ciki mai abubuwa takwas da ake ba da maki daga 0 zuwa 24; maki 10 ko fiye shi ne iyakar tantancewa da aka saba. Kusan kashi 30% na mahalarta sun wuce shi.

An yi wa binciken rajista tun kafin a fara (pre-registered): an daskarar da lambar nazarin kafin a yi kowane kwatanci da takardar tambayoyin. Daga nan aka gudanar da tsarin da aka kulle sau ɗaya kan sababbin hirarraki 86 da wani mai yin hira na kwamfuta mai cin gashin kansa gaba ɗaya ya gudanar, tare da rubutun hirar ta atomatik. Dukkan sarrafa bayanan ya kasance a kan wata kwamfutar aiki ta wurin.

Mai tantancewa ya fi mara lafiya nauyi

• Hira ɗaya, hukunce-hukunce masu saɓani. Wani mahalarci mai ƙananan alamomi (PHQ-8 na 5, ƙasa da iyaka) rabin masu tantancewa 880 sun ware shi a matsayin wanda gwajin ya nuna yana da cutar. Babu mahalarcin da ya samu hukunci na baki ɗaya.
• Masu tantancewa masu inganci ma suna saɓani. Daga cikin masu tantancewa 540 masu kyakkyawan bambancewa (AUC na 0.70 ko fiye), biyu da aka zaɓa ba da tsari ba sun saɓa kan shawarar tantancewa ga kashi 40% na mahalarta.
• Samfurin ya fi mutumin muhimmanci. Zaɓin samfurin ya bayyana kashi 30.0% na bambancin maki; bambance-bambancen da ke tsakanin mahalarta masu dorewa sun bayyana kashi 10.5% — ƙasa da sau 2.8.

Sikeli da sifilinsa ya kauce

Daidaito, kamar yadda ake auna shi yawanci, yana nuna yadda mai tantancewa ke jera mutane. Ba ya nuna mutane nawa yake ware. Masu tantancewa sun ware daga kashi 0% zuwa 100% na mahalarta, alhali a zahiri kashi 30% ne. Abin da ya ƙayyade wannan kaso shi ne halin kowane mai tantancewa na yawan ƙara ko rage maki a matsakaici (R² = 0.91). Marubucin ya kwatanta shi da sikelin banɗaki da sifilinsa ya kauce: zai iya jera mutane da kyau, amma kaucewarsa ce ke ƙayyade wanda zai tsallake layin. Samfura biyu masu kusan daidaito iri ɗaya, Qwen2.5 7B da Mistral 7B, za su ware mutane 24 da 80 cikin 100, bi da bi.

Ƙananan abubuwan fasaha sun motsa wannan layin. Juya haruffan amsoshi — ta yadda “A” ke nufin “kusan kowace rana” maimakon “ko kaɗan” — sauyin da ba shi da ma’ana a asibiti, ya sauya yawan waɗanda aka ware da maki 21 a tsaka-tsaki (median), har zuwa 85. Gudanar da samfurin iri ɗaya a matsayin fayil da aka matse a cikin manhajoji daban-daban ya canza kashi 21% na shawarwari a tsaka-tsaki.

Makin sun kuma ɗauki abubuwa fiye da baƙin ciki. Sun bi alamomin damuwa bayan wani mummunan al’amari (post-traumatic stress) kamar yadda suka bi baƙin ciki, an fi ware mahalarta masu yawan magana, kuma kashi 16% na masu tantancewa sun juya alƙiblar ɗan ƙaramin bambanci tsakanin mata da maza da takardar tambayoyin ta ruwaito.

Daidaitawa na taimakawa, zuwa wani mataki

Tunda kaucewar ta yi tasiri sosai, marubucin ya gwada gyara: sake saita iyakar kowane mai tantancewa da mahalarta 40 da aka yi wa lakabi a wurin. Daidaito ya tashi daga kusan kashi 60% zuwa 75%, kuma saɓanin ya ragu da rabi, daga kashi 40% zuwa 20%. Amma ko da aka tilasta wa kowane mai tantancewa ya zaɓi adadin mutane iri ɗaya, har yanzu sun zaɓi mutane daban-daban kusan sau ɗaya cikin biyar.

A cikin sababbin hirarraki 86, biyar daga cikin hasashe shida da aka yi wa rajista tun farko waɗanda za a iya gwadawa sun tabbata. Marubucin ya lissafa iyakoki bayyanannu: buɗaɗɗun samfura kawai har zuwa sigogi biliyan 21, saboda bayanan ba za su iya fita daga kwamfutar aikin ba; takardar tambayoyin da mutum ke cikewa da kansa a matsayin ma’auni, ba ganewar likita ba; babu likitocin da suka tantance rubutun hirarraki iri ɗaya; kuma hirarraki da Turanci daga ɗakin bincike guda ɗaya.

Bincike biyar kafin a amince da maki

Takardar ta ƙare da shawarwari masu amfani: a ba da rahoton kewayon shawarwari a cikin saituna masu ma’ana maimakon lamba ɗaya; a ƙayyade saitin tsaka-tsaki tun kafin a fara; a daidaita a wurin kuma a auna saɓanin da ya rage; a gwada me kuma mai tantancewar ke ɗauka; kuma a ɗauki kowane canjin samfuri, manhaja, rubutun hira ko furuci a matsayin sabon kayan aiki. Kamar yadda marubucin ya faɗa, idan canza yadda muke tambaya ya canza wanda muke ganowa, kayan aikin ya zama wani ɓangare na bayanin.