Paper Type
ERF
Abstract
Digital humanitarian systems increasingly integrate AI-driven conversational agents to support crisis response, yet how such systems shape prosocial behavior remains poorly understood. In particular, prior work treats compassion as a uniform construct, overlooking how distinct compassion framings influence behavioral outcomes. This study investigates whether and how different forms of compassion embedded within large language model powered conversational agents (LLM-CAs) nudge prosocial behavior. Drawing on four theoretically grounded compassion types, proximal, distal, universal, and relative, we plan to instruction tune LLM-CAs to express these types as well as a non compassionate control. We will conduct a controlled pretest posttest experiment with over 500 participants in a simulated hurricane crisis scenario. Participants will interact with an assigned LLM-CA and make behavioral choices, including donations, engaging in digital volunteerism, and sharing crisis information. By examining the behavioral effects of compassion framings, this study aims to advance the crisis informatics literature.
Paper Number
1134
Recommended Citation
Patel, Hrishitva; Bhattacharyya, Samadrita; Bose, Indranil; and Rao, H. Raghav, "Compassion in Crisis: Nudging Prosocial Behavior Through LLM Conversational Agents" (2026). AMCIS 2026 Proceedings. 3.
https://aisel.aisnet.org/amcis2026/conftheme/conftheme/3
Compassion in Crisis: Nudging Prosocial Behavior Through LLM Conversational Agents
Digital humanitarian systems increasingly integrate AI-driven conversational agents to support crisis response, yet how such systems shape prosocial behavior remains poorly understood. In particular, prior work treats compassion as a uniform construct, overlooking how distinct compassion framings influence behavioral outcomes. This study investigates whether and how different forms of compassion embedded within large language model powered conversational agents (LLM-CAs) nudge prosocial behavior. Drawing on four theoretically grounded compassion types, proximal, distal, universal, and relative, we plan to instruction tune LLM-CAs to express these types as well as a non compassionate control. We will conduct a controlled pretest posttest experiment with over 500 participants in a simulated hurricane crisis scenario. Participants will interact with an assigned LLM-CA and make behavioral choices, including donations, engaging in digital volunteerism, and sharing crisis information. By examining the behavioral effects of compassion framings, this study aims to advance the crisis informatics literature.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
NEXTTRANS