Exploring Generative Models with Middle School Students

Title of the Paper

Exploring the Learning Process of Generative Models for Middle School Students

Paper Information

• Subject Area: Education on generative models, including Generative Adversarial Networks (GANs), artificial intelligence education, and its ethical implications
• Keywords: Generative machine learning, AI education, Generative Adversarial Networks, artificial intelligence, middle school education, applications of generative models, deepfake, AI ethics, game-based learning, cognitive construction

Research Background and Issues

• Identified Challenges:

  • Middle school students are primary users and consumers of generative machine learning technologies, with applications including deepfake content on social media and content generation tools (e.g., FaceApp). However, their understanding of these technologies is limited.
  • There is a lack of courses and frameworks in schools and society that allow middle and high school students to learn about complex AI technologies such as Generative Adversarial Networks (GANs).

• Significance:

  • Generative models (e.g., GANs) have deeply influenced social and technological domains, including image generation, artistic creation, medical imaging, and more, making it essential for future talent to develop the ability to collaborate with AI.
  • Given the ethical risks posed by generative models, such as media manipulation and privacy breaches, middle school students need to learn to critically evaluate AI-generated content and understand its potential societal impacts.

• Research Motivation and Related Work:

  • Existing AI education focuses on foundational algorithms or advanced learners but lacks non-technical, low-threshold teaching materials on generative models suitable for middle school students.
  • This study aims to fill this gap by introducing the concept of "generative AI" into middle school education, equipping students with the ability to understand the technology and its societal implications.

Proposed Solution

• Proposed Solution:

  • Design a learning trajectory (LT) tailored for middle school students, using interactive activities and courses to help them understand the following topics:
    1. The concept of generation;
    2. The structure and working principles of GANs;
    3. Practical applications of GANs in fields such as art, music, and deepfake technology;
    4. Creating and reflecting on content using generative AI tools.

• Innovations:

  • Integrating technical and social ethical discussions ("integrated ethical approach") by embedding ethical issues of generative models into the learning of technology and applications.
  • Designing teaching activities and tools that are user-friendly and require no technical prerequisites.
  • Enhancing the learning experience through real-world case studies and interactive activities, including role-playing games as "generator" and "discriminator."

• Implementation Steps and Key Techniques:

  1. Introduction to the Concept of Generation: Engage students in a "GAN-generated or not" game to help them distinguish AI-generated content.
  2. Teaching GAN Principles: Create a game simulating the "generator vs. discriminator" dynamic, where students take on the roles of generator and discriminator to understand the iterative learning process of GANs through interactive feedback.
  3. Exploring GAN Applications: Use interactive online tools (e.g., AI Duet, Sketch-RNN) to explore the innovative capabilities of GANs in art, music, and facial generation.
  4. Deepfake Video Examination Activity: Teach students to identify deepfake videos, learn methods for detecting algorithmically manipulated content, and discuss ethical implications.
  5. Creative Use of Generative Tools: Have students create stories using generative AI tools while reflecting on the purpose and potential impact of the generated content.

Research Outcomes

• Specific Outcomes:

  • Technical Understanding: Students gained a clearer understanding of technical concepts such as the working principles of GANs and the roles of generators and discriminators.
  • Application Awareness: Students expressed clear opinions on the potential benefits (e.g., artistic innovation) and risks (e.g., privacy invasion, forgery) of AI tools in generating content.
  • Ethical Reflection: Students were able to critically reflect on the ethical issues surrounding generative AI in society and question the authenticity and potential manipulation of content.
  • Effectiveness of Interactive Activities: Various activities facilitated students' integrated understanding of technology, applications, and their underlying ethical considerations.

• Comparison with Existing Solutions:

  • Added a multi-layered learning approach that integrates social and ethical issues into the traditionally linear learning frameworks of mathematics and computer science courses.
  • Focused on a low-threshold generative AI education framework for students aged 11-14, avoiding reliance on advanced mathematical formulas or programming skills.

• Experimental or Evaluation Results:

  • In conceptual understanding tests, students' accuracy in answering GAN-related questions improved significantly from 45.16% to 83.87%.
  • In GAN application learning activities, 88% of students correctly identified the goals of the generator tools, and 60% successfully recognized the datasets used by the discriminator.
  • In the deepfake identification task, despite its technical difficulty, students were able to clearly articulate their observations about deepfake content.

• Limitations and Future Directions:

  • Students exhibited confusion regarding the role of the GAN discriminator and the datasets used in audio-visual generation, indicating a need for enhanced learning objectives on neural networks and training datasets.
  • The design of comprehensive test questions with false options is lacking.
  • Future work should focus on improving the transparency of GAN tools and enhancing the intuitiveness and coherence of teaching activities.