Abstract

Loneliness and social disconnection are recognized risk factors for morbidity and mortality, yet the psychophysiological pathways remain contested. This dissertation examines how social factors relate to vagally mediated parasympathetic activity indexed by heart rate variability (HRV) and tests whether conversational interventions with a human or a large-language-model (LLM)-based companion can acutely reduce loneliness and enhance connection. Across two university sites (BYU, UCSD), N=297 participants (18-27 years) were randomized to (a) a reciprocal self-disclosure conversation with a trained human confederate, (b) a parallel, vocalized LLM chatbot conversation, or (c) a non-social, vocal reading control. Resting HRV (RMSSD, SDNN, HF-HRV) was recorded for 5 minutes, followed by a 20-minute condition and a 5-minute post window; HRV was processed in Kubios HRV Scientific according to best practice. Trait loneliness models (H1/H6) did not yield consistent associations between chronic loneliness and resting HRV. Site and respiration emerged as reliable predictors of resting HRV. Psychological outcomes (H3/H4) showed that human interaction reduced state loneliness and increased connection, synchrony, and feelings of closeness, while the chatbot produced modest reductions in loneliness but weaker relational synchrony and feelings of closeness relative to human interaction. Critically, pre-post (recovery) HRV (H2) showed significant change across all three conditions, in the reverse direction of the preregistered prediction. In the human and chatbot conditions, rather than increasing, recovery HRV decreased from baseline to post-interaction (indicative of parasympathetic withdrawal). The control condition revealed an increase in HRV recovery. Trajectory analyses examining HRV reactivity across six time points throughout the experiment revealed a more nuanced relationship among conditions, variables, and HRV. The most salient finding was the initial increase in parasympathetic response in all three HRV indices during the intervention in the chatbot condition. Moderation analyses (H5) demonstrated that chronic loneliness significantly shaped the magnitude of this phasic change across conditions, indicating systematic differences in autonomic reactivity by loneliness level rather than a uniform pattern, suggesting a blunting effect or physiological thick skin for those experiencing higher loneliness, specifically in the romantic domain. Together, the results sharpen theoretical distinctions between physiological effort and psychological reward in early-stage social engagement, what we frame as "social activation energy." Human conversation uniquely fostered relational synchrony and closeness despite short-term autonomic effort, whereas the LLM yielded modest psychological benefits with comparatively weaker relational signals. Implications include calibrating expectations for acute autonomic responses during connection-building, leveraging scalable AI tools as complements (not substitutes) for human relationships, and prioritizing follow-up on recovery dynamics, longitudinal trajectories of romantic loneliness, and domain-specific effects in diverse samples.

Degree

PhD

College and Department

Family, Home, and Social Sciences; Psychology

Rights

https://lib.byu.edu/about/copyright/