Artificial Intelligence in Secondary Education: Educational Affordances and Constraints of ChatGPT-4o Use

1. Problem Statement and Research Gap

The study addresses a critical gap in the existing literature regarding the application of Generative AI in education. While previous systematic reviews indicate that AI (specifically ChatGPT) can reshape the educational landscape, the research landscape suffers from three specific limitations:

• Version Obsolescence: Most existing studies focus on ChatGPT-3.5, neglecting the newer ChatGPT-4o (released May 2024), which offers enhanced multimodal capabilities (e.g., DALL-E 2.0 image generation, file summarization) without a subscription.
• Educational Level Bias: The majority of AI in education research targets higher education, leaving secondary education (specifically 16–17-year-olds) under-researched.
• Lack of Affordance Analysis: There is a scarcity of empirical studies examining the specific educational affordances (perceived capabilities and advantages) and constraints of GPT-4o from the student perspective.

Research Questions:

1. What are the affordances of GPT-4o for teaching and learning?
2. What constraints limit the use of GPT-4o in teaching and learning?
3. Would secondary education students choose to use GPT-4o in the future?

2. Methodology

The study employed a qualitative research design utilizing a mixed-methods approach (intervention + open-ended questionnaire).

• Participants: A convenience sample of 45 students (2nd year of General Lyceum, ages 16–17) from three classes in Attica, Greece. The demographic consisted of 26 males (57.78%) and 19 females (42.22%). Only 18 participants had prior AI experience.
• Intervention Protocol:
  • Duration: Conducted in May 2024 over a single session.
  • Stage 1 (45 mins): Introductory presentation on AI history, evolution, and applications.
  • Stage 2 (45 mins): Guided creation of GPT-4o accounts and completion of six specific activities to test capabilities:
    1. Information retrieval on a topic of interest.
    2. Creation of a personalized summer job resume based on hobbies.
    3. Image generation using DALL-E 2.0.
    4. Content Summarization: Uploading a 500-page PDF and requesting a summary.
    5. Quiz Generation & Error Detection: Creating a 5-question quiz on a difficult subject and asking GPT to answer it (demonstrating potential hallucinations/errors).
    6. Adaptive Explanation: Asking GPT to explain a complex concept (e.g., quadratic equations) to three different age groups (7, 10, and 14 years old).
  • Stage 3 (15 mins): Free exploration of GPT-4o.
  • Data Collection: An anonymous online questionnaire (Google Forms) with five open-ended questions regarding affordances, future usage intentions, and constraints.
• Data Analysis: Two ICT specialists performed thematic analysis based on Corbin and Strauss's (2014) grounded theory approach, utilizing open, axial, and selective coding.

3. Key Contributions

This study makes three primary contributions to the field of AI in Education:

1. First Empirical Study on GPT-4o in Greek Secondary Education: It is one of the first studies to utilize the specific capabilities of the 4o model (multimodality, file upload) with high school students.
2. Identification of Novel Affordances: While some affordances (skill development, speed) are known, this study identified three new educational affordances not previously reported in the literature:
   • Creation of new knowledge by leveraging prior knowledge (students using AI to build on what they already know rather than starting from scratch).
   • Immediate content feedback (specifically regarding text refinement and summarization).
   • Friendly/familiar interaction environment (the chat interface mimicking peer-to-peer messaging).
3. Shift in Student Perception: The study documents a significant shift in student trust. Before the intervention, most students believed GPT-4o was infallible; after experiencing the "quiz error" activity, they became critically aware of potential inaccuracies.

4. Key Results

A. Educational Affordances (Benefits)

Five distinct categories of affordances were identified:

1. Creation of New Knowledge via Prior Knowledge (N=33): Students recognized using AI to discuss topics they partially knew, leveraging the tool to expand existing mental models.
2. Immediate Content Feedback (N=28): Students valued the instant iterative feedback on their writing and the speed of summarizing long documents (e.g., summarizing a 500-page book in 2 minutes).
3. Friendly/Familiar Interaction (N=25): The messaging interface was perceived as non-intimidating and familiar, similar to chatting with friends, lowering the barrier to entry.
4. Ease and Speed of Information Access (N=18): Rapid response times were cited as a critical factor; slow responses were noted as a potential deterrent.
5. Skill Development (N=11): Students reported potential improvements in digital literacy, computational thinking, problem-solving, and critical thinking, alongside subject-specific mastery.

B. Constraints (Limitations)

Three main constraints were identified:

1. Uncertainty Regarding Truthfulness and Quality (N=16): The most significant concern. Students realized the need to verify AI outputs, leading to a "verification loop" where they must cross-check AI data with internet sources.
2. Anxiety Regarding AI Use (N=8): Two sources of anxiety were noted:
   • Perfectionism: Fear that their own writing would never satisfy the AI's constant suggestions for improvement.
   • Intimidation: Anxiety about interacting with an "entity that knows everything."
3. Privacy Concerns (N=5): Worries regarding data storage, who can access their queries, and the reluctance to provide real names during registration.

C. Future Intent

Despite the constraints, the majority of students (N=42) expressed a positive attitude toward using GPT-4o in future teaching and learning scenarios.

5. Significance and Implications

• Pedagogical Shift: The study suggests that AI should not be viewed merely as a search engine but as a tool for iterative learning and scaffolding (e.g., explaining concepts at different reading levels).
• Critical Digital Literacy: The findings highlight the necessity of teaching students to critically evaluate AI outputs, as the "hallucination" risk is a primary barrier to trust.
• Curriculum Design: Educators can leverage the "friendly" nature of chatbots to increase student engagement but must address the anxiety and privacy concerns through transparent policies and guided usage.
• Future Research: The authors recommend expanding studies to larger samples, different educational levels, and comparing various models (e.g., DeepSeek, Gemini, Claude) to determine if these affordances are model-specific or universal.

Conclusion: The study concludes that secondary education students are generally positive about integrating GPT-4o into their learning, provided that issues of content reliability, anxiety, and privacy are addressed through structured educational interventions.