Exploring AI Problem Formulation with Children via Teachable Machines

Title of the Paper

Exploring AI Problem Formulation with Children through Teachable Machines

Document Information

• Subject Areas: Artificial Intelligence, Child-Participatory Design, Educational Technology
• Keywords: Participatory Machine Learning, Machine Teaching, Values, Design Metaphors, Collaborative Inquiry

Research Background and Problem Statement

• Problems and Challenges:

  • The widespread adoption of artificial intelligence (AI) requires people to better understand problem formulation from an early stage. However, empirical research on how to effectively guide this process in the context of child participation remains limited.
  • In current educational activities, children generally lack the ability to actively control problem formulation, often restricted to modifying the input or output of predefined machine learning models.

• Research Significance:

  • Cultivating children's ability to formulate AI problems is considered a core aspect of AI literacy, which can also help them understand AI's capabilities, limitations, and potential impacts.

• Research Motivation:

  • The authors believe that combining teachable machines and participatory design can provide children with new opportunities to take a more active role in problem formulation.
  • The study aims to explore how to design frameworks and practices to effectively guide children in engaging in such AI problem formulation activities.

Solution

• Methodology:

  • An improved "Big Paper" storyboard method was adopted to structurally guide children in problem formulation.
  • The approach integrates problem reduction heuristics, breaking the design into the following modules: inputs and outputs, technical choices, real-world application scenarios, training processes, error prediction, and coping strategies.

• Innovations:

  • The use of participatory design methods allows children (aged 8-13) to learn AI problem formulation through hands-on practice and collaboration.
  • The study introduces the use of design metaphors (e.g., "intelligent agents" and "super tools") and human values (e.g., capability, responsibility, family safety) as analytical tools to understand and classify children's AI designs.

• Implementation Steps:

  1. Children first use preset teachable machines (e.g., Google Teachable Machine) to familiarize themselves with machine learning and the teaching process.
  2. Using the storyboard tool, children collaborate with adults to complete AI problem formulation, including defining problem scenarios, envisioning machine functions, and planning remedies for failures.
  3. The designed storyboards are analyzed using design metaphors and value models to extract key findings.

Research Findings

• Specific Outcomes:

  • A structured participatory design method was provided to guide children in AI problem formulation.
  • Empirical studies revealed that children often draw inspiration from personal life experiences. Their designs typically assume systems with voice and video functionalities and include multiple error-handling methods.
  • The primary human values reflected in children's designs include capability, responsibility, logical reasoning, and preferences for "family safety" and "comfortable living."

• Comparison with Existing Solutions:

  • Compared to previous activities, this approach emphasizes children's initiative and creativity, allowing for greater flexibility beyond traditional classification tasks.
  • Children's designs tend to align with innovation goals that prioritize human control rather than overly anthropomorphic or fully autonomous AI systems.

• Experimental and Evaluation Results:

  • Children were able to construct complex AI problem formulations through storyboards, including input/output settings, teaching process designs, and error recovery strategies.
  • In terms of metaphor and value analysis, children's designs predominantly aligned with "super tools" or "control center" metaphors, prioritizing human collaboration and intervention.

• Limitations and Future Directions:

  • Limitations:

    • The study's sample size was small, involving only 10 children, which limits the generalizability of the results.
    • The 1:1 collaboration model between children and adults may face challenges in broader practical applications.
    • Some activities may not be fully adaptable for children with visual impairments or other disabilities.

  • Future Directions:

    • Expand the study to include children from diverse cultural backgrounds and age groups.
    • Explore how these learning activities can be integrated into formal educational curricula.
    • Conduct further research to investigate the potential applications of design metaphors and human values in other AI user groups, such as product managers and user experience designers.