Empathy-Aware Large Language Models for AI-Mediated Gambling Addiction Interventions
Session Title
Poster session
Presentation Type
Poster Presentation
Start Date
27-5-2026 12:00 AM
Abstract
Gambling addiction remains a pressing public health challenge, yet many individuals facing problematic gambling behavior avoid traditional counseling due to stigma, accessibility barriers, or personal reluctance. Increasingly, they turn to conversational artificial intelligence (AI) for immediate and anonymous support. While large language models (LLMs) can deliver information, interventions, and coping strategies, their limited ability to express empathy and provide emotional support represents a critical and underexplored gap in AI-mediated addiction consulting. Motivated by this gap, this study investigates the following research questions: (RQ1) How can empathic capabilities be systematically embedded into LLMs for gambling addiction consulting? (RQ2) To what extent does empathic AI improve the effectiveness of addiction-related consulting outcomes? (RQ3) How does empathy-enhanced AI affect users’ emotional responses, conversational satisfaction, and perceived support compared to intervention-only AI? Drawing on theories of empathy, emotion contagion, and affective activation, we design and train models using curated consulting dialogues that integrate not only evidence-based interventions (e.g., cognitive behavioral strategies) but also empathetic responses tailored to users’ emotional needs. The approach focuses on identifying and addressing users’ empathy needs, providing emotional validation, and fostering conversational satisfaction, alongside practical guidance.
Empathy-Aware Large Language Models for AI-Mediated Gambling Addiction Interventions
Gambling addiction remains a pressing public health challenge, yet many individuals facing problematic gambling behavior avoid traditional counseling due to stigma, accessibility barriers, or personal reluctance. Increasingly, they turn to conversational artificial intelligence (AI) for immediate and anonymous support. While large language models (LLMs) can deliver information, interventions, and coping strategies, their limited ability to express empathy and provide emotional support represents a critical and underexplored gap in AI-mediated addiction consulting. Motivated by this gap, this study investigates the following research questions: (RQ1) How can empathic capabilities be systematically embedded into LLMs for gambling addiction consulting? (RQ2) To what extent does empathic AI improve the effectiveness of addiction-related consulting outcomes? (RQ3) How does empathy-enhanced AI affect users’ emotional responses, conversational satisfaction, and perceived support compared to intervention-only AI? Drawing on theories of empathy, emotion contagion, and affective activation, we design and train models using curated consulting dialogues that integrate not only evidence-based interventions (e.g., cognitive behavioral strategies) but also empathetic responses tailored to users’ emotional needs. The approach focuses on identifying and addressing users’ empathy needs, providing emotional validation, and fostering conversational satisfaction, alongside practical guidance.