- The paper demonstrates that agent-mediated intergenerational information sharing enhances older adults’ engagement, as shown by a median of 0 missed responses in the Sharing condition.
- The evaluation combining rule-based dialogue management and qualitative IOS assessments reveals that structured sharing increases perceived relational closeness between grandparents and grandchildren.
- The study highlights practical implications for mitigating elder isolation, advocating for longer interventions and improved privacy controls in dialogue-based systems.
Introduction
The paper "Dialogue Agents that Share Family Information to Strengthen Grandparent-Grandchild Relationships" (2604.12310) presents an empirical investigation into the efficacy of a dialogue agent designed to share personal information between older adults and their grandchildren. The study is motivated by the critical problem of social isolation among older adults, particularly in Japan, and leverages a family-oriented information-sharing chatbot as an intervention. The proposed system aims to enhance not only the user-agent relationship but also intergenerational family bonds through structured, daily conversational exchanges mediated by information transfer.
System Architecture and Design
The proposed agent is implemented as a smartphone chatbot via the LINE messaging platform. The architecture integrates a rule-based dialog manager, an LLM-based response generator (OpenAI API), sentiment and entity analysis (Google NLP API), and persistent user-specific dialogue state (SQL). The dialogue design is agent-initiated, aligning with the user's routine and spanning topics such as sleep, meals, daily activities, and plans for the following day. Dialogue is structured into five-turn micro-conversations to minimize cognitive load and promote engagement.
A distinguishing feature is the conditional inclusion of "Sharing info" responses, wherein content provided by one participant (either grandparent or grandchild) is referenced during conversations with the other. A probabilistic gating mechanism governs when such sharing occurs to prevent repetitiveness and maintain dialogue diversity. The system is explicitly informed by prior research on social curiosity, intimacy, and the differential psychological valuation of intergenerational family ties.
Empirical Evaluation
The core contribution is a ten-day field experiment involving 52 grandparent-grandchild pairs. The experimental design employs a between-subjects comparison of a Sharing condition (information transfer enabled) versus a Non-sharing control (conventional agent-only interaction). Quantitative and qualitative assessment methods are utilized, including:
- Intention to Use scale for subjective willingness to interact.
- Behavioral measures: average response time and frequency of agent-issued reminders.
- IOS (Inclusion of Other in Self) scale for perceived relationship closeness.
- HAD (Hospital Anxiety and Depression) scale for mental health.
- Coded open-ended responses for nuanced evaluation of relational and psychological changes.
Data screening for satisficing ensures the reliability of self-report measures, while logged interaction statistics provide additional behavioral validation.
Key Findings
Willingness to Interact
Older adults in the Sharing condition exhibited significantly fewer missed responses, as indexed by a decreased number of agent reminders (Mdn = 0.00 in Sharing vs. 1.00 in Non-sharing, U = 212, p = .013), indicating elevated behavioral engagement. However, subjective measures (Intention to Use) did not significantly differ between conditions for either group. No significant engagement effects were observed for grandchildren.
Perceived Relationship Closeness
While IOS scale data did not reveal significant condition effects, open-ended responses showed that both grandparents and grandchildren in the Sharing condition more frequently reported positive changes in relational closeness (x²(2) = 9.010, p = .011 for older adults, x²(2) = 8.339, p = .015 for grandchildren). These qualitative findings suggest that information sharing via the agent led to an increased subjective sense of familial connection in both generations.
Mental Health Outcomes
Both cohorts (older adults and grandchildren) demonstrated significant reductions in anxiety scores across the interaction period, with no significant main effect of the Sharing condition and no observable differential effect on depression scores. Qualitative analysis of open-ended mental health responses revealed positive affect in both conditions, but the distribution of coded responses did not significantly differ.
Theoretical and Practical Implications
This study empirically supports the proposition that agent-mediated intergenerational information sharing can enhance behavioral engagement among older adults and strengthen subjective perceptions of familial connectedness. The asymmetry in motivational response—robust for older adults but not for grandchildren—reflects intergenerational differences in the psychological salience of family information, corroborating the theoretical literature on the unique role of grandchildren in elderly adults' social and existential frameworks.
The lack of significant differences in quantitative relationship and mental health scales is noteworthy. It points to the challenges of eliciting measurable psychological change over short intervention periods, as well as potential scale sensitivity deficiencies. The qualitative results underscore the importance of nuanced, open-ended assessment methods when evaluating relational technologies.
The findings provide further evidence for the need to avoid over-reliance on agents as a primary social outlet, an issue increasingly highlighted in recent HAI research reporting adverse long-term outcomes of agent-exclusive socialization [Fang et al., 2025]. The dual-agent strategy—supporting both human-agent and human-human connections—emerges as a more comprehensive approach for mitigating social isolation among the elderly.
Limitations and Future Directions
There are several methodological and design limitations. The intervention period was limited to ten days, insufficient for observing long-chain effects on relationship strength or depression trajectories. The dialogic content was restricted to lightweight, non-intimate topics due to usability and risk considerations, potentially constraining relational impact. Privacy management and user control over sharable content remain under-explored, especially as increased topic depth is considered for future deployments. The study population is demographically and culturally specific (Japanese users with high smartphone penetration), which may limit cross-cultural generalizability.
For future developments, longitudinal studies spanning several months or more are required to elucidate the sustained impact of agent-mediated information sharing on familial relationships and mental health metrics. Enhanced dialogue enrichment, dynamically adjusting topic depth in response to user cues while ensuring privacy and transparency, is an important technical challenge. Incorporating more granular user consent frameworks and explainable sharing mechanisms will be pivotal for wider, real-world adoption.
Conclusion
This work advances the design and empirical evaluation of dialogue agents as active facilitators of intergenerational connectedness. By incorporating selective information sharing, the agent succeeded in increasing behavioral engagement among older adults and subjectively improving perceptions of grandparent-grandchild relationships. The absence of major effects on mental health and quantitative closeness measures during the short trial highlights the complexity of psychological outcomes and the limitations of brief interventions. Future research must address long-term trajectories, topic depth, privacy, and cross-demographic applicability to realize the full potential of such systems for supporting aging populations and familial structures.
