Designing Responsible GenAI Learning Activities in Cognelo: Operationalizing a Design-Space Taxonomy for Programming Education

Context

Generative AI is rapidly transforming programming education. Students can now use tools such as ChatGPT or GitHub Copilot to obtain explanations, generate code, correct errors, receive feedback, or produce complete solutions. These uses offer important opportunities to support learning, but they also raise pedagogical risks: bypassing cognitive effort, becoming dependent on AI, reducing learner autonomy, making it difficult to verify students’ actual understanding, and limiting the traceability of AI contributions.

This project is situated within the context of Cognelo, a modular platform for learning, assessment, and intelligent tutoring centered on pedagogical activities. Cognelo aims to support the design, deployment, and analysis of learning activities, particularly in programming education. The platform can integrate different types of activities, such as multiple-choice questions, Parsons problems, programming exercises, and web programming exercises. It therefore provides a relevant environment for designing, testing, and comparing activities that integrate generative AI in a controlled way.

The project builds on a design-space taxonomy for responsible generative AI use in programming education. In this perspective, responsible use is not defined simply as a matter of allowing or banning AI. Rather, it is understood as a property of the pedagogical configuration. An AI-integrated activity can be considered responsible to the extent that it preserves three dimensions: student agency, the cognitive demand required for learning, and learner accountability for the work produced.

Problem Statement

Many instructors wish to integrate generative AI into their courses but have few concrete tools to translate general principles — such as “using AI responsibly” — into precise, observable, and assessable learning activities. The same technology can produce very different pedagogical effects depending on when AI is introduced into the activity, what type of output it provides, and which mechanisms of verification, justification, or traceability are required from students.

For example, asking AI to generate a complete solution before any student attempt creates a high risk of substituting the student’s reasoning. In contrast, asking students to first produce their own attempt and then use AI to receive feedback, compare solutions, or critique generated code may better preserve cognitive effort and promote reflection.

The central question of the project is therefore:

How can programming activities be designed, integrated, and evaluated in Cognelo in order to operationalize responsible generative AI use?

General Objective

The general objective of this project is to design and evaluate a set of pedagogical activities integrating generative AI in Cognelo, based on a design-space taxonomy of responsible GenAI use in programming education.

Specific Objectives

1. Formalize a design model for responsible learning activities
   Define a design grid that positions an activity according to two main dimensions:
   • the timing of AI entry in the activity: before, during, or after the student’s independent effort;
   • the scope of the AI output: complete solution, partial solution, feedback/evaluation, or material to critique.
2. Design activity patterns that can be integrated into Cognelo
   Develop several models of responsible pedagogical activities, for example:
   • post-attempt activity with AI feedback;
   • debugging activity supported by hints;
   • critique activity based on AI-generated code;
   • comparison activity between a student solution and an AI suggestion;
   • activity with a justification and traceability log of AI use.
3. Implement a prototype in Cognelo
   Adapt or develop a Cognelo module that enables the creation, configuration, and deployment of these activities. The prototype should make it possible to control the conditions of AI use, for example:
   • access to AI only after an initial attempt;
   • limitation of AI output to hints or feedback;
   • requirement for students to justify which AI suggestions they accept or reject;
   • preservation of traces of AI interactions.
4. Evaluate the activities with students or instructors
   Conduct an exploratory study to evaluate the relevance, usability, and pedagogical value of the designed activities. The evaluation may focus on:
   • students’ perceptions;
   • instructors’ perceptions;
   • quality of student productions;
   • students’ ability to explain their work;
   • interaction traces collected in Cognelo.
5. Analyze how different configurations influence the level of pedagogical responsibility
   Compare different activity configurations in order to examine which ones seem to better preserve cognitive effort, student autonomy, and learner accountability for the work produced.

Possible Research Questions

The project could be guided by the following research questions:

1. How can a taxonomy of responsible generative AI use be translated into concrete pedagogical activity patterns in Cognelo?
2. Which configurations — post-attempt feedback, AI-based critique, generated-code analysis, graduated hints — are perceived as the most responsible and useful by instructors and students?
3. To what extent do activities that include sequencing mechanisms, limits on AI output, justification requirements, and traceability preserve student agency, cognitive demand, and learner accountability?
4. Which traces collected in Cognelo can be used to assess the pedagogical responsibility of an AI-integrated learning activity?

Proposed Methodology

The project could be structured in four phases.

Phase 1 — Conceptual Analysis and Design of Activity Patterns

The student will begin by analyzing the taxonomy of responsible GenAI use and identifying the pedagogical configurations most relevant to Cognelo. This phase will lead to the production of a design grid that links each activity to its design choices: timing of AI entry, scope of AI output, verification mechanisms, level of traceability, and expected role of the student.

Phase 2 — Prototype Development in Cognelo

The student will develop or adapt a Cognelo module that allows the creation of activities integrating generative AI according to explicit pedagogical constraints. The prototype could include:

• an activity editor for instructors;
• configuration of the moment when AI becomes accessible;
• control over the type of AI output allowed;
• a student interface for requesting feedback or critiquing an AI output;
• a mechanism for justifying decisions made by the student;
• a trace system for analyzing AI use.

Phase 3 — Exploratory Deployment

The prototype may be tested in a real or semi-real programming education context. Depending on the constraints of the project, the study may take the form of:

• a classroom experiment;
• a pilot study with a small group of volunteer students;
• an evaluation by instructors;
• a comparison of pedagogical scenarios without full classroom deployment.

Phase 4 — Data Analysis

The data analyzed may include:

• activity traces in Cognelo;
• student productions;
• responses to justification questions;
• interactions with AI;
• perception questionnaires;
• short interviews with students or instructors.

The analysis will aim to determine whether the designed activities effectively support responsible AI use, and which design dimensions appear to be the most important.

Expected Contributions

This project is expected to produce three types of contributions.

Scientific Contribution

The project will contribute to the empirical operationalization of a design-space taxonomy for responsible generative AI use in programming education. It will help move from a conceptual framework to concrete, implemented, and evaluable activities.

Technological Contribution

The project will enrich Cognelo with a module or set of features for designing responsible AI-integrated activities. These features may serve as a basis for other research projects on intelligent tutoring, formative assessment, automated feedback, and learning analytics.

Pedagogical Contribution

The project will provide instructors with reusable models of learning activities that integrate generative AI in a structured way, rather than simply allowing students to use external tools freely. These models may help instructors move from a binary logic of “AI allowed or forbidden” toward a logic of controlled pedagogical design.

Expected Deliverables

The expected deliverables of the project may include:

1. a targeted literature review on responsible generative AI use in programming education;
2. a design grid for responsible learning activities;
3. a set of pedagogical activity patterns;
4. a prototype integrated into Cognelo;
5. an exploratory study with students or instructors;
6. a master’s thesis;
7. ideally, a scientific paper submitted to a conference or journal in educational technology, AIED, or computing education.

Examples of Activities to Be Designed

Activity 1 — AI Feedback After Student Attempt

The student must first submit an initial solution to a programming exercise. AI then becomes available to provide feedback, without giving a complete solution. The student must indicate which suggestions they accept, which ones they reject, and why.

Activity 2 — Critique of AI-Generated Code

Cognelo presents AI-generated code containing errors or questionable design choices. The student must test, critique, correct, and justify their modifications. In this case, AI is not used as an answer provider, but as an object of analysis.

Activity 3 — Debugging with Graduated Hints

The student works on an incorrect program. AI can provide progressive hints, but it cannot directly provide the corrected code. The activity aims to preserve the debugging effort while reducing frustration.

Activity 4 — Comparison Between Student Solution and AI Suggestion

After proposing their own solution, the student receives an AI-generated suggestion. They must compare the two approaches, identify differences, explain which one is preferable, and justify their decision.

Activity 5 — Traceability Log of AI Use

The student must document each interaction with AI: what was asked, what was proposed, what was accepted or rejected, and how this influenced the final solution.