Kwenzekeni
Abacwaningi uMargaret Mitchell, Avijit Ghosh kanye no-Samir Passi baphikisa ephepheni lesikhundla ukuthi imiklamo yamanje ye-ejenti ye-AI ayikusekeli ngokwanele ukugada komuntu. Bathi ukuthembela okwandisiwe ezinhlelweni ezizenzakalelayo nakho kungalulaza amakhono ababonisi abawadingayo ukuze basebenzise ukwahlulela okubucayi.
Iphepha le-arXiv elithunyelwe ngo-Aug. 24, 2026, lithi umqondo ojwayelekile wokugcina umuntu “ku-loop” akuyona, ngokwawo, impendulo enokwethenjelwa ezingozini zama-agent e-AI akhula ngokuzilawula. Ababhali, uMargaret Mitchell, Avijit Ghosh kanye noSamir Passi, bachaza umsebenzi wabo njengephepha lesikhundla esikhundleni sombiko wemodeli entsha ye-AI, ukuthunyelwa komkhiqizo noma ukuhlolwa okulawulwayo.
Isimangalo sayo esimaphakathi siwukuthi izindlela zamanje zokuthuthukisa nokuthumela ama-agent e-AI zingabukela phansi izimo ezidingekayo ukuze kuqashwe abantu ngempumelelo. Ababhali bahlonza izinkinga ezimbili ezihlobene. Okokuqala, baphikisa ngokuthi indlela ama-agent e-AI aklanywe ngayo ingakhinyabeza ukugada okusebenzayo. Umbonisi angase agcine igunya ngokobuchwepheshe kuyilapho engenalo ulwazi, ukulawula noma ithuba elingokoqobo elidingekayo lokuhlola ukuthi i-ejenti yenzani. Okwesibili, baphikisa ngokuthi ukusetshenziswa okwandisiwe kwe-automation kungehlisa amandla okuqonda adingekayo ukuze kuqashwe. Kuhlaka lwephepha, i-automation ayishintshi nje ukusatshalaliswa komsebenzi; kungase futhi kwenze buthaka amakhono omuntu uhlelo lokwengamela elicabanga ukuthi azohlala etholakala.
Iphepha lixhuma umsebenzi wokuzenzakalela kanye nokusebenzisana kwekhompyutha yomuntu nezinqubo ze-AI-ejenti. Idinga amandla okusebenza ezingeni lokuklama kanye nezimiso zenhlangano ezihloselwe ukusekela ukwahlulela okubucayi kwabaphathi kanye nokulwisana ne-atrophy yamakhono ehambisana nokusetshenziswa okuzenzakalelayo isikhathi eside. I-abstract ayikubali ngokuningiliziwe lezo zindleko noma izivumelwano, ngakho izindlela zazo ezithile kanye nezidingo zokuqalisa azikwazi ukutholakala embhalweni womthombo onikeziwe. Isincomo esibanzi sababhali ukuthi izidingo zabantu kufanele ziphathwe njengezibaluleke njengamandla e-ejenti lapho amasistimu esethuthukisiwe.
Baxwayisa ngokuthi ngaphandle kokusekelwa okucacile kwezidingo zomqondo zokusebenzisana ne-ejenti yabantu, ama-agent e-AI angase agqugquzele ngokungenzi lutho ukuwohloka kwamakhono abantu lapho ukuphepha nokuziphendulela kwabo kuncike khona. Leyo yingxabano emayelana nokwakhiwa kwesistimu kanye nokubusa, akubona ubufakazi bokuthi yonke i-ejenti ye-AI yamanje isivele ibaxoshile abenzi bezinqumo ezingabantu noma ibangele ukwehla okulinganisekayo kwengqondo.
Imininingwane yomthombo: arxiv.org ↗
Kungani kubalulekile
Iphepha lisusa ukunaka kokuthi ngabe umuntu ukhona yini ngokusemthethweni kuye ekutheni lowo muntu angakwazi yini ukuqonda, inselele futhi angenele ezenzweni ze-ejenti. Ukungqubuzana kwayo kunemithelela ekwakhiweni nasekusetshenzisweni kwezinhlelo ze-AI ezisetshenziswa ngegunya elibalulekile.
Iphuzu elibaluleke kakhulu lephepha ukuthi ukubandakanyeka komuntu okuzisholo kungase kungalingani nokulawula komuntu okunenjongo. Umuntu owabelwe ukuqapha i-ejenti ye-AI angase angakwazi ukuyigada kahle uma isistimu yenza izenzo zayo zibe nzima ukuyiqonda, yethula ulwazi kabi, ikhawulela ukungenelela, noma ikhuthaza ukugunyazwa okujwayelekile. Umthombo awusho ukuthi noma iyiphi i-interface eyodwa noma ukusetshenziswa kukhiqize lokhu kwehluleka, kodwa ikhomba umbuzo osebenzayo wokuphepha: yini okufanele umuntu akwazi futhi ayenze ukuze ukwengamela kube okwangempela kunokuthi kube yinqubo?
I-agumenti iphinde igqamise ingozi engabonakali kakhulu yokuzenzakalela. Uma abantu behlehlisa ngokuphindaphindiwe izincomo ezizenzakalelayo noma beyeka ukwenza amakhono ayisisekelo, ikhono labo lokubona amaphutha, ukubuza imibuzo ecatshangwayo nokwenza ngokuzimela lingancipha. Kubenzeli be-AI abangakwazi ukwenza imisebenzi yezinyathelo eziningi noma basebenze ngokuzimela okwandayo, lokho kungenzeka kubalulekile ngoba ukungenelela kuvame ukwenzeka kuphela lapho kukhona okungahambanga kahle. Ngakho-ke isistimu yokwengamela ingase ingathembeki ngokunembile njengoba i-ejenti ithola umthwalo owengeziwe.
Lolu hlaka lubalulekile ezinhlanganweni ezinquma ukuthi lingakanani igunya lokunikeza ama-agent e-AI. Iphakamisa ukuthi ukubuyekezwa kokuthunyelwa akufanele kuhlole kuphela ukunemba kwemodeli noma ukuqedwa komsebenzi, kodwa futhi nendima yomuntu ohlelweni: noma ngabe ababonisi bathola umongo owanele, ukuthi bangakwazi yini ukubekela inselele noma bamise izenzo, kanye nokuthi ubuchwepheshe babo buyagcinwa yini ngokuhamba kwesikhathi. Lokhu kuyimiphumela yempikiswano yababhali, hhayi okutholakele ukuthi iphepha libonisa kuyo yonke imboni noma inani labantu.
Leli phepha futhi liwusizo kakhulu ngoba liphatha amandla omuntu njengengxenye yecala lokuphepha lezinhlelo ze-agent ye-AI. Uma i-ejenti incike ekwahluleleni komuntu, khona-ke ukulondoloza lokho kwahlulela kuba yimfuneko yesistimu kunokuba kube udaba lokuqeqeshwa noma lwabasebenzi olungezwe ngemva kokuthunyelwa. Umthombo awuqinisekisi ukuthi indlela ehlongozwayo izosebenza, kodwa yenza indaba yokuhlola ukwengamela njengekhono elisebenzayo okufanele liklanywe, lisekelwe futhi linakekelwe.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Ongakubuka ngokulandelayo
Imibuzo ebalulekile iwukuthi ingabe ukukhokhela kwedizayini okuhlongozwayo kanye nezivumelwano zenhlangano zithuthukisa ukugada komhlaba wangempela, indlela i-atrophy yekhono okufanele ilinganiswe ngayo, nokuthi ingabe onjiniyela nabaphakeli bayazisebenzisa yini izivikelo eziphatha izidingo zombonisi njengemfuneko yesistimu ewumongo.
Umbuzo osheshayo ngowokuthi lisho ukuthini leli phepha eligcwele ngokuthi "izindleko zezinga lokuklama." Umthombo onikeziwe uzikhomba njengezindlela ezihloselwe ukusiza ababonisi basebenzise ukwahlulela okubucayi, kodwa awucacisi uhlobo lwabo, ukuthi bazosebenzisana kanjani nokuzimela komenzeli, noma ibuphi ubufakazi obungabonisa ukuthi bayasebenza. Leyo mininingwane izonquma ukuthi isiphakamiso singaqondisa umkhuba wobunjiniyela noma sihlala siwuhlaka lomqondo.
Umbuzo wesibili ukuthi i-atrophy yekhono kufanele ilinganiswe kanjani. Ukugomela okungaqondakali kokuthi ukusetshenziswa okuzenzakalelayo okunwetshiwe kungehlisa izinga lamakhono okuqonda adingekayo ukuze kugadwe, kodwa akunikezeli isilinganiso sobuningi, inani locwaningo, umkhathizwe noma iqembu lokuqhathanisa. Ucwaningo olwengeziwe luzodinga ukucacisa ukuthi yimaphi amakhono asengozini, ngaphansi kwaziphi izimo, nokuthi ukuqeqeshwa, ukushintshashintsha komsebenzi, ukuzijwayeza ngamabomu noma ezinye izinyathelo zenhlangano zingawalondoloza.
Ukuhlola kufanele futhi kudlulele ngale kokuthi i-ejenti iyawuqeda yini umsebenzi wayo eyabelwe. Amasistimu asebenza kahle kumamethrikhi emisebenzi asengadala izinkinga zokugada uma efihla ukungaqiniseki, enze ukungenelela kubize noma akhuthaze abantu ukuthi bagunyaze okuphumayo ngaphandle kokubuyekezwa okuzimele. Umthombo usekela ukubuka ukuhlola okuhlola ukusebenzisana komenzeli womuntu kanye nemiphumela yomhlaba wangempela, nakuba wona ngokwawo ungakubiki ukuhlaziya okunjalo.
Okokugcina, abaqambi bazodinga ukunquma ukuthi isibopho sikuphi lapho ukwengamela kwenziwa buthaka ukwakheka kwesistimu noma ukwenza kwenhlangano. Ababhali banxusa onjiniyela nabaqambi ukuthi basebenzise izincomo zabo noma izindlela ezifanayo, kodwa umthombo awucacisi amazinga omthetho, izimfuneko zokulawula, izilawuli eziqondene nomkhakha othile noma izindlela zokuziphendulela. Futhi ayiqinisekisi ukuthi ingabe i-agumenti isebenza ngokulinganayo kuwo wonke ama-ejenti ahlukene wezakhiwo, imisebenzi noma amazinga okuzimela. Lokhu okungaziwa kufanele kuhlale kusobala njengoba kudingidwa isiphakamiso futhi sivivinywa.