Generative AI as a code tutor: catalyst for autonomy or trap of dependence?

Virtual Generation

Serigne Fallou Niang, Ndeye Sophie Seck, Fatou Bintou Ba ,Mame Abdoulaye Ndiaye, UN-CHK (Cheikh Hamidou Kane Digital University)

Training / education / pedagogy

At UN-CHK (Cheikh Hamidou Kane Digital University) , generative AI as a tutor is transforming distance learning, shifting between empowerment and passivity. Faced with isolation and a lack of support, a programming student uses assistants (ChatGPT, Copilot) to overcome obstacles. Our project proposes a structured pedagogical use of AI: it doesn't provide the solution, but guides the learning process. This intelligent tutoring strengthens autonomy and ensures active progress for the learner.

This situation deserves our attention because it represents a tipping point: AI is no longer simply a research tool, but a cognitive partner that is redefining the act of learning. Challenges and tensions: Autonomy vs. Dependence: The challenge is to maintain the necessary cognitive effort. If AI does too much of the work, the student develops an illusion of knowledge without real mastery. Reliability and hallucinations: A tutor who confidently makes mistakes can mislead the learner in a lasting way, especially in asynchronous contexts without immediate human verification. Learning and cognition: Opportunity: Reduced costs for students who do not have access to expensive private tutors. Risk: A "two-tiered education" where some benefit from human teachers and others only from automated interfaces. Social relationships and student life: The risk of isolation increases. If AI answers everything, students will seek out their peers or teachers less, weakening the learning community. Teaching: The role of the teacher is evolving from "transmitter of knowledge" to "architect of learning paths" and mediator of AI.

Our team is eager to tackle this challenge because we are witnessing a looming educational divide: on one hand, students who are enhancing their skills with AI; on the other, those who are using it as a crutch, losing all ability to solve complex problems. We hope to transform AI from a "solution generator" into a "cognitive mirror." This process will allow us to: Better understand: the precise moment when AI assistance shifts from facilitation (relieving mental workload) to atrophy (stifling critical thinking). Question: the relevance of current assessment methods in computer science. If AI can write code, what should we actually be teaching and grading? Transform: the student's role, so that they move from passive code consumers to critical architects capable of validating and optimizing what AI proposes.

To transform AI from a "crutch" into a "catalyst for autonomy" in coding, our team identifies these types of critical resources: Cognitive science experts: to understand "mental load" and define the point at which AI assistance hinders memorization and logical understanding. Senior/professional developers: to identify the critical skills that AI cannot (yet) replace (architecture, security, code ethics) and that students must master independently. Instructional designers: to help us transform a traditional coding course into an "AI-assisted code review" learning path. Access to large language models (LLMs): using models like GPT-4 or Claude 3.5 Sonnet to test system prompts that force AI to remain in a tutoring role (never provide the complete code, respond with hints). Integrated Development Environments (IDEs): a sandbox environment where students code and AI intervenes only under specific conditions. Partnership with a coding school or bootcamp: to conduct a comparative study (Group A: free AI vs. Group B: "restrictive tutor" AI). Learning logs: access to anonymized chat histories to analyze when a student abandons their independent thinking to copy and paste a solution. Digital ethics experts: to oversee the protection of student data and ensure that the algorithm does not favor certain coding styles over others. The ultimate goal is to create an AI tutoring protocol that could be adopted by universities.