AI chatbots should be bridges, not destinations
BMJ · 03.03.2026 · 392 · s398
DOI: 10.1136/bmj.s398 ↗Kurz gefasst
Leserbrief in der Zeitschrift BMJ, online erschienen am 03.03.2026. Verfasst von Kazumi Kubota. Beteiligte Einrichtung: Shimonoseki City University, Research Organization, Shimonoseki, Japan. Die Arbeit nennt 5 Literaturangaben.
Abstract (Originalsprache)
Shelmerdine and Nour describe how AI chatbots might widen access to support while also risking psychological and social harms, including deepening isolation. 1 The key question for health systems is not whether chatbots can simulate empathy, but whether they increase offline connection—especially for older adults, in whom loneliness often coexists with disability and practical barriers to leaving home. 2 3 If a chatbot is optimised for “time on screen,” it may reduce the feeling of loneliness without reducing loneliness itself. A more defensible goal is to treat chatbots as bridges rather than destinations. For commissioners and regulators, this implies a simple requirement: evidence of a net increase in human-to-humancontact.Anyimprovementinlonelinessscoreswithreducedface-to-faceinteractionshould betreatedasharm,notbenefit.Evaluationshouldshiftawayfromengagementmetricstowardsoutcomes thatreflectsocialreintegration.“Timeincommunity”canbemeasuredusinglowburden,verifiable endpoints—forexample,completionofascheduledlinkworkercontact,attendanceatacommunityactivity, orrepeatparticipationat30or90days—ratherthanusingconversationlengthormessagevolume. A bridge model also requires human infrastructure. Older adults’ isolation is often maintained by concrete constraints (mobility, home hazards, unmet needs for assistive equipment, transport, and fragmented social care). 3 Conversationalsupportcannotinstallagrabbar,removeafallhazard,orcoordinatepracticalservices. Without an accountable human pathway that can act on identified needs, chatbot “support” risks becoming compassion without traction. Social prescribing provides a pragmatic backbone. An AI enabled workflow could summarise needs, match local resources, and draft structured referral prompts, but must be coupled to human coordination. Finally, engagementsignals(escalatingsessionlength,repeatedhopelessness,sleepdeterioration)shouldbetreated assafetyinputstriggeringreviewandescalation—notassuccessmetrics.Ethically,chatbotsshouldhelp peoplereturntooneanother,safelyandmeasurably,ratherthansubstituteforhumancontact
Autor:innen und Institutionen
- Shimonoseki City University, Research Organization, Shimonoseki, Japan.
Zitieren
APA
Kubota, K. (2026). AI chatbots should be bridges, not destinations. BMJ, 392, s398. https://doi.org/10.1136/bmj.s398
Vancouver
Kubota K. AI chatbots should be bridges, not destinations. BMJ. 2026;392:s398. doi:10.1136/bmj.s398
BibTeX
@article{Kubota2026,
author = {Kubota, Kazumi},
title = {AI chatbots should be bridges, not destinations},
journal = {BMJ},
year = {2026},
volume = {392},
pages = {s398},
doi = {10.1136/bmj.s398}
}
EUVIMED