Care-Based Eco-Feedback Augmented with Generative AI: Fostering Pro-Environmental Behavior through Emotional Attachment

Title of the Paper

Care-Based Eco-Feedback Augmented with Generative AI: Fostering Pro-Environmental Behavior through Emotional Attachment

Paper Information

• Research Domain: Sustainable HCI, data-driven environmental behavior change
• Keywords: eco-feedback, care-based intervention, generative AI, conversational interaction, gamification, emotional attachment

Research Background and Problem

• Observation and Problem: The authors highlight that traditional eco-feedback systems often rely on data-driven methods to raise users' awareness of energy consumption. However, these approaches tend to overlook emotional connections and are less effective for groups lacking environmental awareness.
• Significance: As the global energy crisis intensifies, reducing energy consumption is crucial for achieving net-zero emission targets. However, traditional feedback methods may fail to inspire long-term behavioral changes, necessitating new approaches that cater to a broader audience.
• Research Motivation: Addressing the empirical research gap on "how emotional attachment influences energy-saving behavior"; exploring interactive designs powered by generative AI to enhance user motivation for behavioral change.

Solution

• Methods and Solution:
  A novel care-based eco-feedback system is proposed, centered around a digital character—a virtual pet named Infi—whose vitality is influenced by the user's energy-saving behavior. The system integrates generative AI technology to support emotional interactions between users and the virtual character, employing a Tamagotchi-like experience to engage users.
• Innovations:
  1. For the first time, generative AI conversational technology is integrated into care-based designs to strengthen emotional bonds between users and the system.
  2. A pathway model is constructed and empirically validated, linking motivational components (system design) to behavioral outcomes (energy-saving behavior).
  3. Differential effects of the design are analyzed for user groups with varying levels of environmental awareness.
• Implementation Steps and Key Technologies:
  1. Design Principles: Based on Ping Zhang's user motivation design principles, including autonomy, feedback mechanisms, social connection, group influence, and emotional triggers.
  2. Specific Design:
     • The system connects to users' energy-saving data via smart meters.
     • User interactions include answering energy-saving questionnaires and chatting with Infi. Each interaction enhances Infi's "vitality" or provides visual feedback on improved energy-saving levels.
  3. Application of Generative Conversational Technology: OpenAI's GPT-3.5 API is utilized to support two-way dialogues, creating engaging and educational interactions.

Research Outcomes

• Specific Findings:

  1. System Effectiveness Validation: Randomized controlled experiments and subsequent surveys reveal that AI-interactive care-based designs significantly enhance users' emotional attachment and increase motivation for energy-saving behavior.
  2. Differentiated User Performance: Among users with varying levels of environmental awareness, emotional attachment has a stronger impact on energy-saving behavior for those with low environmental awareness.
  3. Sustainability Testing: In a three-month follow-up study, the system effectively improved self-reported energy-saving behaviors.

• Advantages Compared to Existing Solutions:

  1. Strengthening emotional connections between users and the system significantly boosts energy-saving behavior, beyond mere information-driven approaches.
  2. Design strategies targeting users with low environmental awareness address shortcomings of traditional feedback methods.
  3. Generative conversational interactions provide flexibility and depth, making feedback more personalized.

• Experimental and Evaluation Results:

  1. In a randomized controlled experiment with 720 participants, the emotional attachment mean score for users in the GenAI version group was 4.51 (significantly higher than the control group's 3.17).
  2. The system increased both self-reported intentions for energy-saving behavior and actual behavioral changes, such as donations or reduced energy consumption.

• Limitations and Future Directions:

  1. Geographical Constraints: The experiment was conducted only in the U.S.; further validation across diverse cultural contexts is needed.
  2. Technical Challenges: System design needs optimization to address technical issues arising from reliance on GenAI models, such as occasional message truncation or repetitive content generation during chats.
  3. Psychological Backlash: A minority of users exhibited resistance to emotionally driven feedback. Future efforts should focus on mitigating negative ecological anxiety or other psychological barriers.

Research Contributions and Application Recommendations

1. Encourages designers to incorporate emotional attachment as an agency element in eco-feedback designs, such as using virtual characters to foster a sense of care.
2. Advocates for balancing emotional experiences with behavior-oriented designs to avoid potential negative psychological effects.
3. Suggests future research to explore the profound impact of other design elements (e.g., personalization or sense of ownership) on behavioral change.