OceanChat: The Effect of Virtual Conversational AI Agents on Sustainable Attitude and Behavior Change

Paper Title

OceanChat: The Effect of Virtual Conversational AI Agents on Sustainable Attitude and Behavior Change

Publication Info

• Topic area: The use of conversational AI agents to promote pro-environmental behavior and attitudes.
• Keywords: conversational AI, pro-environmental behavior, marine conservation, sustainability, large language models, environmental education, anthropomorphism, empathy, behavior change, interactive systems.

Background and Problem

• Problem / challenge: Traditional environmental education struggles to translate awareness into meaningful pro-environmental actions. Existing technological interventions often rely on static information, require specialized hardware, or lack interactivity, limiting their accessibility and engagement potential.
• Significance: Addressing climate change and plastic pollution in marine ecosystems is critical for biodiversity and human health. Innovative, scalable solutions are needed to foster emotional resonance and actionable behavior change.
• Motivation and related work: Prior research has explored eco-feedback systems, gamified interfaces, and virtual reality for environmental engagement but has not fully leveraged the potential of conversational AI combined with emotionally resonant character design. This paper builds on these efforts by integrating large language models with species-specific, interactive narratives.

Solution

• Proposed approach: OceanChat, an interactive system featuring AI-driven marine characters (beluga whale, jellyfish, seahorse) designed to promote pro-environmental behavior through dynamic, personalized dialogue.
• Novelty:
  1. Open, scalable architecture for environmental communication using conversational AI.
  2. Empirical insights into how species representation and interaction modalities influence environmental attitudes and behaviors.
  3. Design principles for creating emotionally resonant, AI-generated environmental characters.
  4. Introduction of a synthetic pre-evaluation methodology using large language models to refine system design.
• Procedure and key techniques:
  • Developed three experimental conditions: static scientific information, static character narratives, and interactive conversational narratives.
  • Conducted a randomized controlled trial (N=900) to evaluate the system’s impact on environmental beliefs, behavioral intentions, and sustainable choices.
  • Integrated real-time dialogue, species-specific narratives, and behavioral psychology principles into the AI system.
  • Used synthetic participants for pre-deployment testing and refinement.

Results

• Concrete findings:
  • Conversational Character Narrative condition significantly improved sustainable choice preferences (β = 0.342, p = 0.031) and pro-environmental intentions (β = 0.173, p < 0.001).
  • Emotional connection (e.g., empathy with marine animals, β = 0.404, p < 0.001) was a key driver of behavioral outcomes.
  • The beluga whale character elicited the strongest emotional engagement, outperforming jellyfish and seahorse conditions in perceived likeability (β = 0.361, p < 0.001) and empathy (β = 0.184, p = 0.040).
• Advantage over baselines:
  • Interactive dialogue outperformed static scientific information and static character narratives in fostering behavioral intentions and sustainable product preferences.
  • Static approaches were less effective in creating emotional resonance and behavioral commitment.
• Experiments / evaluation:
  • Conducted a between-subjects experiment with three conditions and three species, analyzing outcomes such as climate change beliefs, sustainable consumption, and information sharing.
  • Used validated scales and regression models to measure environmental attitudes, behavioral intentions, and perceptions of AI characters.
• Limitations and future work:
  • Limited impact on deeper psychological measures like psychological distance and climate policy adoption.
  • Short-term study duration; long-term effects remain untested.
  • Residual timing differences across conditions and reliance on pre-existing attitudes.
  • Future work should explore repeated interactions, species-specific linguistic cues, and multi-sensory modalities to deepen engagement.

Summary

OceanChat demonstrates the potential of conversational AI agents to foster pro-environmental behavior and emotional connection with marine life. The system’s interactive, species-specific narratives significantly improved sustainable intentions and choices, with emotional resonance playing a critical mediating role. While the beluga whale character achieved the strongest engagement, the study highlights the importance of balancing anthropomorphism with species authenticity. Limitations include the short-term nature of the intervention and its limited impact on deeper psychological and policy-related outcomes. Future research should focus on long-term engagement strategies, expanded species-specific cues, and scalable deployment in real-world contexts.