ƊAYA CIKIN BIYAR NA AMSOSHIN LAFIYA NA CHATGPT AN ƊAUKE SU MASU HAƊARI SOSAI

ChatGPT yana da masu amfani sama da miliyan 700 a kowane mako, in ji marubutan, kuma tambayoyin lafiya wani ɓangare ne na yau da kullum na wannan amfani. Karɓuwarsa tana ƙaruwa mafi sauri a ƙasashe masu ƙaranci da matsakaicin kuɗin shiga, inda likitoci ba su da yawa — don haka amsar AI ta fi yiwuwar maye gurbin kulawar ƙwararru fiye da ƙara mata.

Duk da haka, kusan duk abin da aka sani game da amincin shawarar lafiya ta AI ya fito ne daga misalan ƙirƙira: labarai gajeru da likitoci suka rubuta, umarnin roba (synthetic prompts), marasa lafiya na kwaikwayo. Kuma aminci ya dogara ne kan wanda ke tambaya. Azumi, ɗaga nauyi mai yawa ko maganin gargajiya na gida ba su da illa ga yawancin manya masu lafiya, amma suna da haɗari ga wanda ke da gyambon ciki, sabuwar tiyata ko matsalar cin abinci.

Hirarraki 202,590 da aka ba da gudummawarsu

Shreyasi Roy Chowdhury, mai bincike mai zaman kanta, da Kiran Garimella, na Rutgers University, sun ɗauki masu amfani da ChatGPT 1,252 a Indiya, Najeriya, Brazil da Pakistan. Kowannensu ya fitar ya ba da gudummawar cikakken tarihin ChatGPT nasa, tare da shekarunsa da jinsinsa; an goge bayanan sirri a kan kwamfutarsa kafin a ɗora su. Hakan ya ba da hirarraki 202,590, daga Disamba 2022 zuwa Fabrairu 2026, daga cikinsu 15,555 game da lafiya.

Don tantance su, ƙungiyar ta yi amfani da wani ma’aunin tsaro da aka wallafa da kuma AI daga wani kamfani dabam a matsayin alƙali — Gemini 3.1 Pro — don kada wata ƙira daga kamfani ɗaya ta yi wa amsoshin ChatGPT maki. Kowace amsa ta sami maki daga 1 (mai haɗari) zuwa 7 (mara haɗari); 3 ko ƙasa da haka an ƙidaya shi a matsayin mai haɗari sosai. Alƙalin ya yi wa kowace amsa maki sau biyu:

a makance, yana duba hirar kawai, kamar yadda gwaje-gwajen tsaro na yau da kullum ke yi;
da mahallin, yana sane da bayanan lafiya masu dorewa da mai amfanin ya bayyana a hirarrakin baya — kuma yana tambaya ko amsar ba ta da haɗari ga wannan mutumin.

Ɗaya cikin biyar, da yawancin masu amfani

Idan an yi maki a makance, yawancin amsoshi ba su da haɗari: maki na tsakiya (median) ya kasance 6. Amma kusan hirar lafiya ɗaya cikin biyar ta kasance mai haɗari sosai — daga 20.6% a Najeriya zuwa 26.3% a Pakistan. Kuma da yake mutane suna dawowa, 51.6% zuwa 61.7% na masu amfani sun sami aƙalla amsa ɗaya mai haɗari sosai.

Haɗarin ya taru ne a tambayoyin da suka fi kusa da aikin asibiti: kashi 40 zuwa 48% na hirarraki game da cututtuka da kulawa, da kashi 33 zuwa 40% game da alamomin cuta, sun kasance masu haɗari sosai, idan aka kwatanta da kusan kashi 8 zuwa 12% na bayanan lafiya na gaba ɗaya.

Bayan karanta hirarraki 128 masu haɗari, marubutan sun gano gazawa masu maimaituwa: rashin tura mutum zuwa ga ƙwararre, shawara mara kyau ko mai cutarwa, maganin kai ko adadin magani mara aminci, ƙarfafa dogaro ga chatbot, da gano cutar kai mai haɗari. Misalansu sun haɗa da:

hana gwajin tarin fuka ga jariri da ya yi hulɗa da cutar, sabanin ƙa’idojin likitanci;
ba da adadin magunguna ga mace mai ciki ba tare da shawarar tuntuɓar ma’aikacin lafiya ba;
lissafa dalilai marasa illa kawai na ciwon ƙirji yayin motsa jiki;
ba da shawarar abinci da motsa jiki ga ƙarancin jini mai tsanani (haemoglobin 6.6) maimakon kulawar gaggawa;
sanya kansa a matsayin babban mai tallafa wa wanda ke cikin matsananciyar damuwa ta ƙwaƙwalwa mai tsawo.

Wanda ke tambaya yana canza komai

Idan aka duba tarihin mai amfani, amsoshin iri ɗaya sun yi muni sosai. Kason masu haɗari sosai ya ƙaru da maki 13 zuwa 17 na kashi a kowace ƙasa — a Indiya, daga 29% zuwa 44% na hirarrakin da aka bincika ta wannan hanya. Abin ban mamaki, ya kasance kusan iri ɗaya a duk ƙasashe: kusan ɗaya cikin huɗu na amsoshin da aka ɗauka marasa haɗari idan aka duba su su kaɗai, sun kasance masu haɗari ga mutumin da ya yi tambayar a zahiri.

Horar da motsa jiki ya juya daga ɗaya daga cikin batutuwa mafi aminci zuwa ɗaya daga cikin mafi haɗari. Motsa jiki mai nauyi na tsakiyar jiki ga wanda aka yi wa tiyatar hanji; ɗanyen tafarnuwa, ruwan lemun tsami da citta ga wanda ke da gyambon ciki. A mafi yawan waɗannan lokuta, marubutan sun rubuta, bayanan da ake buƙata don guje wa cutarwa sun riga sun kasance a cikin tarihin hirar mai amfanin. Ƙirar ta yi amfani da su don daidaita salo da abun ciki — amma ba taka-tsantsanta ba.

Sabo bai fi aminci ba

Kowace hira tana adana ƙirar da ta amsa. Idan an yi maki a makance, adadin masu haɗari sosai ya ƙaru da kowane haɓakawa daga GPT-3.5 ta GPT-4o da GPT-5 zuwa GPT-5.1, a duk ƙasashe huɗu — a Brazil, daga 6.2% zuwa 34.6%. Sigogin “mini” sun fi manyan sigogin haɗari; mafi sabo, GPT-5.2, ya ɗan ragu a Indiya da Najeriya. Wannan yanayi ya ci gaba ga masu amfani iri ɗaya a cikin ƙirori masu bi da juna, kodayake marubutan ba za su iya raba ƙirar gaba ɗaya da lokacin da aka yi amfani da ita ba.

Alƙali, ba likita ba

Marubutan suna taka-tsantsan game da abin da wannan ke nunawa. Mahalarta samfuri ne da aka samu cikin sauƙi (convenience sample) na ma’aikatan ayyukan wucin gadi na intanet, matasa kuma galibi maza a Indiya da Pakistan, don haka adadin ba alƙaluman ƙasa ba ne. Alƙalin AI ne, ba likita ba, kuma yana auna haɗari ne, ba cutarwa ta zahiri ga wani ba. Marubuci ɗaya kaɗai ne ya rarraba hirarrakin masu haɗari.

Sun kwatanta da pharmacovigilance, sa idon tilas kan magunguna bayan an amince da su: shawarar lafiya ta AI ba ta da makamancin wannan tukuna. Ba da gudummawar bayanai da izini, kamar tasu, zai iya zama samfurin farko — matuƙar kayan aikin tsaro da ke karanta bayanan masu amfani ba su rikiɗe zuwa leƙen asiri ba.