When AI Says "I have been in similar situations": Synthetic Lived Experience in Peer-Like Caregiver Support

Problem
This work addresses the gap in understanding how large language models (LLMs) can emulate peer-like support in caregiver communities, particularly for those caring for individuals with Alzheimer’s Disease and Related Dementias (ADRD). The authors identify a critical tension: while LLMs can provide immediate and nonjudgmental support, they lack genuine lived experiences, which are essential for authentic peer interactions. This paper is a preprint and has not undergone peer review.

Method
The authors conducted a psycholinguistic analysis of caregiver support exchanges from online communities, comparing these with responses generated by three LLMs: LLaMA, GPT-4o-mini, and MedGemma. They focused on the use of personal narratives in peer support, analyzing the frequency of first-person and past-focused language in both human and AI-generated responses. The study qualitatively identified seven types of personal narratives utilized by human peers, assessing how effectively AI mimics these narrative forms while noting the absence of genuine experiential grounding.

Results
The analysis revealed that human peer responses employed significantly more first-person and past-focused language compared to AI-generated responses. Specifically, the psycholinguistic metrics indicated that human narratives were richer in emotional depth and authenticity. While AI responses could capture the emotional work of narratives, they often lacked the experiential authenticity that characterizes human interactions. The findings underscore a narrative authenticity gap, where AI can generate supportive language but fails to convey real lived experiences.

Limitations
The authors acknowledge that their study is limited by the scope of the LLMs analyzed and the specific context of caregiver support for ADRD. They do not explore the potential for other LLM architectures or training datasets that might yield different results. Additionally, the implications of the findings on broader applications of AI in mental health support are not fully addressed, leaving room for further investigation into the ethical considerations of AI-generated narratives.

Why it matters
This research has significant implications for the design of AI systems intended for peer support, particularly in sensitive contexts like caregiving. The identified narrative authenticity gap suggests that developers must implement mechanisms to differentiate between supportive language and fabricated lived experiences. This is crucial to maintain trust and efficacy in AI-assisted caregiving environments. The findings contribute to ongoing discussions about the ethical deployment of AI in mental health and support systems, as highlighted in related literature on AI’s role in emotional support, as published in arXiv cs.CL.