Source-linked AI summary
When AI "Works," When Does Help Begin?: Intergenerational Support Around Older Adults' LLM Usage
Hyehyun Chu, Yuri Lee, Yeon Su Park, Saelyne Yang, Juho Kim
TL;DR
Intergenerational support around older adults’ LLM use is underexplored, especially how family members help assess risks when operational failures are not visible. Using interviews and scenario-based think-aloud activities with six older adults and seven younger adults, the study finds that partial visibility and general warnings often leave support without reusable calibration knowledge, motivating designs that preserve older-adult control.
Problem
The everyday interactional consequences of family members’ expanded role in helping older adults judge AI uses, disclosures, outputs, and action safety remain underexplored.
Method
The study used semi-structured interviews and scenario-based think-aloud activities with six older adults and seven younger adults.
Results
Partial visibility limited younger adults’ ability to know when support was needed, while warnings often lacked reasons and support rarely accumulated into reusable calibration knowledge.
Takeaways & Limitations
Intergenerational LLM-support systems should close the feedback gap while preserving older adults’ control and autonomy.
Abstract
from arXiv · showhide
LLMs are becoming part of everyday life, including for older adults (OAs). OAs often learn digital technologies with younger family members, who have traditionally served as "warm experts" providing trusted and personalized operational help. LLMs expand this role: family supporters may also help OAs judge appropriate uses, consider what information to disclose, assess the credibility of outputs, and decide when AI-generated advice is safe to act on. We conducted a formative qualitative study with six OAs and seven younger adults (YAs), using semi-structured interviews and scenario-based think-aloud activities. OA participants described using LLMs to lighten their recurring reliance on family, while preserving family as a selectively invoked support channel. However, because LLMs rarely produced visible operational breakdowns, YAs had limited signals for when support was actually needed. Instead, YAs relied on OAs' partial disclosures and negotiated intervention through general warnings and self-imposed action boundaries. As a result, family support often solved an immediate problem without leaving reusable calibration knowledge for future use. Based on these findings, we propose design implications for intergenerational LLM support (e.g., consentful help requests, learning-oriented family support that preserves OA task ownership).
1 Introduction
LLMs extend intergenerational support from operational instruction toward helping older adults assess appropriate uses, disclosure, output credibility, and action safety. A formative study examines how this emerging stewardship is enacted and identifies a feedback problem in family intervention.
- Younger family members traditionally provide trusted, personalized operational help as “warm experts” within ongoing relationships.
- LLMs may shift younger family members toward “warm stewards” who monitor, interpret, and manage AI-related risks while sustaining emotional support.
- Family supporters may help older adults judge appropriate AI uses, decide what information to disclose, verify uncertain outputs, and assess when advice is safe to act on.
- The everyday interactional consequences of this expanded support role remain underexplored, and these judgments can be difficult even for younger helpers.
- The study uses semi-structured interviews and scenario-based think-aloud activities with six older adults and seven younger adults to examine intergenerational LLM support and its challenges.
- Risk warnings often arrived without stated reasons and were read as criticism, while younger adults lacked visibility into later sessions and could not confirm whether interventions changed practices.
2 Methodology
The formative qualitative study recruited six older adults and seven younger adults with experience using or supporting LLM chatbots. Sessions combined interviews, live scenario-based tasks, family-help recall, and thematic analysis.
- Participants: The sample included six older adults aged 60–70 and seven younger adults aged 24–34, recruited through South Korean community forums, welfare centers, and snowballing.
- Participants: Eligibility required older adults to be at least 60, use LLM chatbots, and have experience giving or receiving family-based support related to LLM use.
- Procedure: Each approximately 60-minute session comprised an opening interview, a live scenario-based task, and a family-help recall interview.
- Procedure: Participants discussed AI practices and verification, then addressed health and legal or regulatory concerns while thinking aloud and sharing their screens.
- Procedure: Researchers reconstructed family-help episodes by asking older adults about requesting or resisting help and younger adults about helping, gatekeeping, and responsibility.
- Analysis: Two authors iteratively coded transcripts, reconciled discrepancies, refined the codebook, independently coded remaining transcripts, and resolved disagreements through team meetings.
3 Results
Older adults used LLMs to regulate dependence on family while retaining family as a selective support channel, but preferred learning-oriented help was inconsistently provided. Partial visibility and divergent competence standards made support an unspoken boundary negotiation that rarely accumulated into reusable knowledge.
- 3.1.1 F1: OAs used LLMs to regulate dependence within family relationships.: Older adults used LLMs to strengthen autonomy, handle everyday questions independently, and decide which issues were worth bringing to family.
- 3.1.1 F1: OAs used LLMs to regulate dependence within family relationships.: LLM use also let some older adults display newly acquired competence while preserving family as a safe channel for admitting uncertainty.
- 3.1.2 F2: OAs’ preferred form of help was not always the one provided.: Family support took four forms: doing the task, brief verbal rules, demonstration, and use-case observation.
- 3.1.2 F2: OAs’ preferred form of help was not always the one provided.: Older adults most frequently wanted demonstrations, but the support they received did not consistently take that form.
- 3.2.1 F3: Support rarely accumulated into reusable applicable knowledge.: Doing tasks for older adults could solve immediate problems without providing a reproducible method, whereas demonstrations offered strategies potentially reusable later.
- 3.2.1 F3: Support rarely accumulated into reusable applicable knowledge.: Support episodes often resolved individual problems without accumulating into reusable knowledge for more independent future use.
- 3.2.2 F4: Partial visibility turned support into unspoken boundary work.: Older and younger adults judged competence and risk from partial observations, using different standards for prompting skill versus distinguishing trustworthy answers.
- 3.2.2 F4: Partial visibility turned support into unspoken boundary work.: Families rarely made task boundaries explicit: older adults chose what to show, younger adults filled gaps with assumptions, and interventions arrived as after-the-fact warnings.
4 Design Implications
The paper proposes intergenerational LLM-support designs that make uncertainty and stakes visible, preserve older-adult control, and turn family help into actionable knowledge that can be reused.
- Design implications aim to close the feedback gap while preserving older-adult control and autonomy.
- Systems could surface uncertainty and task stakes because fluent answers can conceal difficulties with utilization and calibration.
- Family support could persist as concrete steps attached to reasons, avoiding warnings that lack specificity and prompt completion that lacks durability.
- Structuring questions upfront can preset evaluation criteria and lower the downstream verification burden.
- Use-case observation most reliably expanded older adults’ range of use, suggesting systems support ambient exposure alongside explicit explanation.
- Systems could preserve older-adult control by letting older adults choose the scope, timing, audience, and form of family support.
A Appendix
The appendix characterizes the six older-adult participants’ LLM usage and the seven younger-adult participants’ family contexts. Together, the tables specify participant demographics, usage patterns, relationships, and contact arrangements.
- Older-adult participants: Table 2 reports the six older-adult participants’ demographic characteristics and self-reported LLM usage patterns.It includes the LLM services used, length of usage experience, and typical use frequency.
- Younger-adult participants: Table 3 reports the seven younger-adult participants’ relationships and family contexts with their older relatives.It identifies matched older-adult participants, older relatives’ ages, living arrangements, and contact frequency.
- Younger-adult participants: The younger-adult table enables comparison of matched participation, relationship, age, living arrangement, and contact frequency across family pairs.A dash in the Matched OA column denotes an older relative who did not participate in the study.