Do LLMs Outperform Human Agents in Anxiety Symptom Alleviation? A Multi-Modal Wearable Sensing Study
Paper Type
Complete
Abstract
Anxiety disorders represent a growing global health challenge, with traditional intervention methods often limited by accessibility and scalability constraints. This study investigates the comparative effectiveness of AI-based conversational agents versus untrained human agents in real-time anxiety intervention using continuous physiological monitoring. We employed an experimental protocol with anxiety induction followed by talk and joke-based interventions, monitoring participants through wearable sensors capturing heart rate, skin conductance response (SCR) frequency, skin temperature, and photoplethysmography signals. Our feasibility analysis identifies heart rate and SCR frequency as the most reliable biomarkers. Surprisingly, AI agents demonstrated superior and more consistent physiological anxiety reduction across heart rate (10+ BPM reduction) and SCR (4-6 peaks/min reduction) compared to untrained human agents, who showed more variable effects. These findings suggest that while human connection remains valuable for mental health support, standardized AI-delivered interventions may offer reliable, scalable solutions for basic anxiety management, particularly as accessible alternatives.
Paper Number
1223
Recommended Citation
Wang, Zi; Zhao, Jiyue; He, Sen; Zhang, Linghan; and Li, Lin, "Do LLMs Outperform Human Agents in Anxiety Symptom Alleviation? A Multi-Modal Wearable Sensing Study" (2026). AMCIS 2026 Proceedings. 3.
https://aisel.aisnet.org/amcis2026/sig_health/sig_health/3
Do LLMs Outperform Human Agents in Anxiety Symptom Alleviation? A Multi-Modal Wearable Sensing Study
Anxiety disorders represent a growing global health challenge, with traditional intervention methods often limited by accessibility and scalability constraints. This study investigates the comparative effectiveness of AI-based conversational agents versus untrained human agents in real-time anxiety intervention using continuous physiological monitoring. We employed an experimental protocol with anxiety induction followed by talk and joke-based interventions, monitoring participants through wearable sensors capturing heart rate, skin conductance response (SCR) frequency, skin temperature, and photoplethysmography signals. Our feasibility analysis identifies heart rate and SCR frequency as the most reliable biomarkers. Surprisingly, AI agents demonstrated superior and more consistent physiological anxiety reduction across heart rate (10+ BPM reduction) and SCR (4-6 peaks/min reduction) compared to untrained human agents, who showed more variable effects. These findings suggest that while human connection remains valuable for mental health support, standardized AI-delivered interventions may offer reliable, scalable solutions for basic anxiety management, particularly as accessible alternatives.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
SIGHealth