"I want it to talk like Darth Vader": Helping Children Construct Creative Self-Efficacy with Generative AI

Title of the Paper

"I want it to talk like Darth Vader": Helping Children Construct Creative Self-Efficacy with Generative AI

Paper Information

• Subject Area: Artificial Intelligence (AI) and Children's Creativity Education
• Keywords: Artificial Intelligence, Creativity, Children, Participatory Design, Co-Design, Constructivism

Research Background and Problem

• Research Questions and Challenges:
  • Current generative AI (GenAI) creativity support tools are often designed with the assumption that users possess a certain level of domain knowledge. However, the significant differences in creative needs between children and adults make existing designs potentially inadequate for child users.
  • Children's creative processes are socially and culturally dependent, and systems lack support for children's unique social contexts and knowledge construction needs.
  • Most AI research focuses on education or skill enhancement rather than supporting creativity or improving children's confidence in their creative abilities (creative self-efficacy).
• Significance:
  • Creativity is crucial for children's development, fostering cognitive growth, social skills, and learning interests. Moreover, as AI tools increasingly permeate the education sector, understanding how they can be effectively applied is critical.
  • AI as a tool to assist children's creativity may contribute to innovative educational methods. However, poorly designed systems could limit children's engagement and creative expression.
• Research Motivation:
  • To explore the potential of generative AI tools in supporting children's creativity.
  • To investigate how generative AI can enhance children's creative self-efficacy, thereby helping them develop creative thinking.

Solution

• Methods and Design:
  • Conducted six participatory design (PD) sessions to study the cognitive and behavioral interactions of children aged 7-13 with generative AI tools, including ChatGPT and DALL-E.
  • Systematically observed children's feelings and behaviors during activities, such as refining prompts, switching tasks, and addressing challenges in generating creative outputs.
• Innovations:
  • Proposed an explanatory model based on four supportive contexts to analyze and evaluate how children interact creatively with generative AI:
    1. System Affordances: Includes the capabilities of AI systems and children's understanding of those capabilities.
    2. Child’s Process: Involves task construction, generating and evaluating outputs, and adapting to challenges flexibly.
    3. Domain Understanding: Relates to expectations and understanding of cultural and community contexts.
    4. Creative Intention: Includes personal interests, motivations, and the task's relevance within the environment.
  • The model emphasizes supporting micro-creative moments (mini-c moments), such as children's satisfaction with outputs and discovering new creative methods through learning tool functionalities.
• Implementation Steps and Key Techniques:
  • Introduced typical generative AI tools like ChatGPT (text creation) and DALL-E (visual creation) in participatory design activities.
  • Designed various creative tasks, such as generating stories, poems, or combined visual-text works.
  • Used reflective discussions, behavioral observation logs, and inductive and thematic data analysis methods to gather research insights.

Research Findings

• Specific Outcomes:
  • Developed an explanatory model centered on "micro-creative moments," summarizing key contexts and dependencies for supporting children's creative interactions with generative AI.
  • Four key findings:
    1. Children need guidance to understand and adapt to the unique functionalities of AI tools.
    2. When AI fails to meet expectations, children adjust by rewriting prompts, providing contextual information, or switching to new topics.
    3. Many AI systems use formal language, which creates cognitive barriers for children lacking domain knowledge.
    4. Children's creative intentions and task environments significantly influence their perceptions and usage of generative AI.
• Advantages Over Existing Tools:
  • Addresses the shortcomings of current AI tools that overlook the needs and behavior patterns of child users, providing a theoretical framework for designing more child-friendly generative AI tools.
  • Proposes viewing generative AI as a constructivist learning tool to support the development of creative self-efficacy, rather than merely a substitute for task execution.
• Experiments and Evaluation:
  • Conducted iterative design experiments to observe children's interaction paths, using a consistent coding system with the research team to provide detailed explanations of obstacles and potentials in the creative process.
• Research Limitations:
  • The sample was limited to 12 children from a single geographic location, and task-setting differences may affect the generalizability of the results.
  • The range of tools studied was limited, including only a few representative tools such as ChatGPT and DALL-E.
• Future Directions:
  • Expand the sample size to study the applicability of generative AI across more cultural and geographic contexts.
  • Explore the deeper impact of domain knowledge depth on children's creative interactions.
  • Optimize and refine interaction designs that support children's creativity, such as introducing age-specific AI interface optimizations and personalized assistance.

Conclusion

This study explores the potential of generative AI to support children's creative tasks, proposing an explanatory model that emphasizes children's creative confidence and learning outcomes. This research framework lays the foundation for applying generative AI tools in human-centered education, highlighting their importance for childhood education and AI literacy development.