Noel: A Chatbot Persona to Support Children Designing for Others

Research Background and Issues

• Discovery of Issues or Challenges: Children do not always have opportunities to interact with the target groups they are designing for in design learning, such as individuals with special needs (e.g., visually impaired individuals). Direct contact with these target users is both challenging and potentially unethical, necessitating alternative methods to help children understand and empathize with users from different backgrounds and with diverse needs.
• Significance: Through scenario-based learning and reflection, students can enhance their empathy and understanding of the specific needs of particular groups. This is crucial for fostering broader social responsibility and design capabilities.
• Research Motivation and Related Work: Previous studies have shown that "personas" or "dynamic persona" models can help designers better understand users. However, traditional persona models are often static and lack interactivity, limiting designers' perception and empathy. Interactive AI-based chatbots have the potential to bridge these gaps.

Solution

• Method/Solution: The authors propose a dynamic chatbot persona named Noel, developed based on the GPT-4 language model, simulating a 12-year-old child with low vision. Noel aims to help students cultivate design empathy through the Design-Based Learning (DBL) process.
• Innovative Features: Unlike traditional static personas, Noel is a dynamic, interactive, and personalized character capable of expressing complex emotions and providing real-time feedback. This "dynamic interaction capability" makes Noel's participation in the students' design process more personalized and vivid.
• Implementation Steps and Key Technologies:
  1. Data Collection: Interviews were conducted with two visually impaired adults and an experienced special education teacher to gather insights into the life experiences and challenges of children with low vision.
  2. Knowledge Base Construction: Based on the interview themes and related resources, a knowledge base was created encompassing the lived experiences of individuals with low vision, forming Noel's backstory.
  3. Technical Implementation: Noel is built on GPT-4 and utilizes Streamlit to provide an interactive interface for real-time interaction with students and to record conversations.
  4. Educational Implementation: During a three-day workshop, students designed solutions tailored to Noel's needs and engaged in multiple rounds of interaction with Noel during the design stages.

Research Outcomes

• Specific Outcomes:
  1. Over the three-day workshop, students completed a full design task following the design thinking process (including empathy, problem definition, ideation, prototyping, and testing).
  2. Noel successfully sparked students' interest and understanding of the lives of children with low vision, while also inspiring their design ideas. For instance, students designed prototypes such as wristbands and marking tapes to assist with object recognition based on Noel's feedback.
• Comparison with Existing Solutions and Advantages:
  1. Compared to static user personas, Noel can provide personalized responses to students' questions in real time, offering critical information and inspiration.
  2. Noel is not merely a "tool"; it exhibits human-like emotional characteristics, further promoting interaction and empathy between students and the target group.
• Experimental or Evaluation Results:
  1. Students demonstrated multidimensional design empathy through Noel, including emotional engagement, reflection on personal experiences, and sensitivity to others' needs.
  2. In surveys, students reported that Noel was most helpful during the design feedback stage but less so during the prototyping stage.
  3. Data analysis showed that students' design empathy increased progressively during interactions, as they displayed greater interest in Noel's life experiences and incorporated his feedback to improve their designs.
• Limitations and Future Directions:
  1. Some of Noel's feedback was perceived by students as repetitive or insufficiently in-depth. Future improvements should enhance the chatbot's contextual memory and precision.
  2. This study did not conduct a comparative evaluation with human interaction counterparts. Future research could include comparative experiments to validate Noel's effectiveness.
  3. Multimodal interactions (e.g., voice and image generation) could be added to further enhance user experience and support for design prototyping.
  4. Long-term impacts, particularly on students' sustained empathy toward target groups, have yet to be evaluated.

In summary, this study demonstrates the significant potential of AI-based dynamic chatbots in education and design practice, particularly in helping students engage with target user groups that are difficult to directly interact with. It also provides clear directions for further improvement and expansion.