Generative AI as a Coding Tutor: Catalyst for Autonomy or Trap of Dependency?

Team Name: Virtual Generation

Use of AI Tools

If you used AI, please specify:

Which tool(s) did you use?

• ChatGPT

For what purpose(s)? (e.g., research, idea structuring, writing, data analysis, image creation.)

• We used AI, particularly kikivoice.ai and ChatGPT, for voice-over correction, improving paragraph structure, and making the content more attractive.

How did these tools contribute to your work?

• These tools helped improve our reflection and our overall vision of the project. They assisted us in structuring our ideas, being more precise in our analyses, and producing clearer and more coherent content.

External Contributions and Feedback

Please indicate the people who supported you during the process (e.g., experts, professors, professionals):

• Name: Dr. Ndeye Massata Ndiaye

• Role: Professor

• Type of contribution: Mentorship

Initial Contribution

Please provide the link (URL) to your initial contribution on the platform: Generative AI as a code tutor: catalyst for autonomy or trap of dependence?

Final Contribution

1. Reflection on the Process and Evolution (Phase 2)

Our project underwent a profound evolution during Phase 2. Initially, our reflection mainly focused on the technical limitations of AI. Interactions with experts and peers helped us understand that the real issue was not restricting the tool itself, but transforming the student’s cognitive posture.

Influence of Expert Feedback

Feedback on the concept of “desirable difficulty” became the trigger for our major innovation. We realized that the immediacy of AI creates a “path of least effort” that is harmful to deep learning. Conferences on human-centered AI, particularly those by Saiph Savage, encouraged us to ensure that humans remain the final arbiters of logic and reasoning.

Modifications Made

• Transition to the “Interrogative Tutor”: We abandoned the idea of a simple filter and adopted a Socratic approach in which AI never directly provides the solution.

• Integration of “Pedagogical Silence”: In response to criticism regarding the atrophy of thinking, we introduced a mandatory response delay to encourage personal research and reflection.

• Automated Human Relay: To prevent frustration from becoming an obstacle, we added an alert mechanism directing the student to a human tutor in cases of prolonged blockage.

Improvement of the Proposal

This process transformed a reactive idea limiting AI into a proactive proposal: creating a “Cognitive Mirror” that strengthens student autonomy.

2. Presentation of the Final Proposal

PROJECT: GENERATIVE AI AS A CODING TUTOR

Catalyst for Autonomy or Trap of Dependency?

I) Situation and Context Studied

At Université Numérique Cheikh Hamidou Kane (UN-CHK), distance learning makes AI particularly useful for reducing student isolation. However, without an appropriate pedagogical framework, it can become a crutch that creates an illusion of competence: the student produces functional code without truly mastering the underlying logic.

Usage Situations (Personas)

• Persona 1: “Fatou the Shortcut Seeker”
  A data science student. Whenever a Python error appears, she copies and pastes it into ChatGPT. She gets the correction, but does not understand the root cause of the error.
  Risk: Illusion of competence.

• Persona 2: “Doudou the Explorer”
  Uses AI to ask: “Explain this loop concept as if I were 10 years old.” He uses AI to understand complex concepts (personalized learning enabled by generative AI), not simply to generate answers.
  Opportunity: Personalized tutoring.

II) Critical Analysis

1) Pedagogical Analysis

The weakening of the process of reaching a solution by clearly defining the necessary steps may lead to the atrophy of algorithmic thinking. Learning to code requires a period of “friction” struggling with the problem. AI removes this friction.

Risks

1. Cognitive Dependency:
   Students may develop the reflex of consulting AI before even attempting to reason independently. This can weaken analytical skills, perseverance in problem-solving, memorization of concepts, and the development of personal problem-solving strategies.

2. Illusion of Competence:
   A student may produce functional code without truly understanding why it works, how it was designed, in which cases it would fail, or how to adapt it deeply.
   This creates a gap between visible performance and real mastery.

3. Weakening of Fundamental Learning:
   If AI intervenes too early or too strongly, it may short-circuit the acquisition of core foundations: syntax, algorithmic logic, structured thinking, manual debugging, and critical code reading.

2) Ethical and Responsibility Issues

• Biases and Hallucinations:
  AI may suggest outdated or insecure methods.
  Students who lack experience may not have enough critical distance to evaluate the response.

• Intellectual Property:
  Using generated code raises questions regarding the originality of submitted work for certification purposes.

3) Ecological and Organizational Issues

• Environmental Impact:
  Each request for help or code correction consumes energy. Multiplied by thousands of students, this usage becomes problematic.

• Structural Challenges:
  Educational institutions must invest in detection tools or, preferably, in new forms of assessment such as oral examinations or in-person exams without internet access.

Generative AI tends to remove the “friction” necessary for anchoring knowledge. Our challenge is therefore to reposition the tool so that it supports cognitive effort instead of replacing it.

III) Analysis of Debates and Collective Arbitrations

1) “Fundamentalist” Position (Rejected): Banning AI During Learning

Argument: The only way to guarantee that students work independently.

Reason for Rejection: Technically unenforceable and disconnected from the professional world, where AI is already becoming a standard tool.

2) Position (Rejected): Total and Free Integration of Copilot

Reason for Rejection:

• Excessive dependence on AI

• Reduced personal reflection effort

• More superficial learning

• Risk of copying code without understanding it

3) “Mediator” Position (Selected): AI as an Interrogative Tutor

Argument: AI should not directly provide the solution, but rather guide the student. The tool is authorized, but a rigorous dialogue method is imposed.

Our objective is to train “Critical Architects.” In this approach, AI becomes a “Cognitive Mirror” reflecting the student’s own reasoning back to them. Competence no longer consists only in mechanically writing code, but also in auditing and questioning the machine.

IV) Major Proposal: The “Reflection Trace” Protocol

Proposed Solutions

1. Frame AI usage with a clear charter

2. Train students in the critical use of AI

3. Reinforce the learning of fundamentals without AI

4. Design assessments that measure real understanding

5. Require traceability of AI usage (such as the “reflection traces” described here)

6. Develop metacognition by encouraging students to ask themselves questions such as:

   • Did I truly understand this solution?

   • Could I reproduce it on my own?

   • What did AI contribute that I would not have found myself?

   • What do I still need to learn?

7. Train teachers in new support postures:

   • Integrating AI into pedagogical scenarios

   • Designing adapted instructions

   • Detecting superficial usage

   • Valuing critical analysis rather than simple code production

8. Reduce inequalities in access by selecting accessible tools, providing complementary local resources such as libraries, and designing instructions that do not disadvantage students with limited access.

Key Features of the Protocol

• Pedagogical Silence (Deliberate Latency):
  A waiting period is imposed before each AI response in order to encourage initial personal reflection and research.

• Interrogative Tutoring:
  AI uses a Socratic method and never directly provides complete code, instead guiding the student through questioning.

• Automated Human Relay:
  If the student remains blocked for several hours despite the help of the interrogative tutor, the system alerts a human teacher and sends them the “Reflection Trace” (history of reasoning and attempts) to enable targeted intervention.

V) Conditions for Implementation

Initial Training (2 hours)

Mandatory module on AI ethics and pedagogy.

Comparative Experimentation

A scientific study comparing:

• Group A: Free AI usage

• Group B: Virtual Generation Protocol

Success would be measured through a final examination conducted without any AI assistance.

Trust Agreement

Signing of a responsible usage charter.

Dedicated Tools

Use of application programming interfaces (APIs) with system instructions such as:
“You are the tutor. Never provide the complete code. Ask questions to guide the student.”

VI) Conclusion

By transforming AI into a demanding pedagogical partner, UN-CHK ensures the intellectual sovereignty and genuine autonomy of its future graduates.

KEY POINTS

• Social Impact: Reduction of online student isolation through the AI tutor.

• Political Impact: Preventing the emergence of a generation of developers dependent on private technologies.

Video Presentation