ARIA : How should higher education institutions structurally evolve their teaching methods and assessment practices in response to AI ?

Team name : ARIA

Section 1 : Use of AI tools :

Tools used

• ChatGPT (OpenAI)

• Claude (Anthropic)

• Gemini (Google)

• DeepL (translation)

Purposes

1. Research support

   Exploring academic literature on higher education reform, AI in pedagogy, and governance frameworks

   Synthesizing keynote inputs and challenge discussions

2. Structuring and refining ideas

   Organizing complex arguments across the three structural axes of our proposal

3. Drafting and reformulation

   Clarifying written arguments in both French and English

   Improving fluency and precision of our final submission

4. Intellectual sparring

   Testing the solidity of our arguments by prompting AI to play devil’s advocate

   Stress-testing our long-horizon framing against counterarguments

How these tools supported our work

AI tools acted mainly as cognitive support tools throughout the process. They helped us organize complex ideas, reformulate our reasoning more clearly, and explore different perspectives related to higher education, governance, AI ethics, and institutional transformation.

The tools did not replace our reflection process but rather supported the articulation and structuring of our arguments. Human judgment remained central in selecting, refining, and validating all final ideas and proposals.

Section 2 : External feedback & contributions :

Prof. Stéphane Frénot — Project tutor, Critical feedback on Phase 1 positioning, guidance to deepen structural analysis.

Caroline Beslin — Associate Lecturer in Project Management and Communication, Feedback on communication clarity, social relevance, and implementation methodology.

Alexandra Pregent — PhD Researcher, Contributions on AI governance, harmful AI uses and ethical frameworks.

Dr. Alain Goudey — Associate Dean for Digital, Distinction between AI use modes, cognitive atrophy, emotional dependency and productivity trade-offs.

Saiph Savage — Assistant Professor, Northeastern University Keynote on human-centered AI : bias, hallucinations, governance, accountability, and designing AI responsive to human values.

Dr Christelle Scharff — Professor of CS, Pace University, Questions on AI adaptation in higher education, MOOCs, AI tutoring, and epistemic quality of AI-generated content.

Cognitive Psychology Master's student (external) — External support / Brainstorming Provided scientific grounding on Social Arousal and Neural Synchrony

Chelza INZOUDDINE — Challenge organizer, INRIA Coordination and contextual information throughout the challenge.

Section 3 : Initial contribution

How should higher education institutions structurally evolve their teaching methods and assessment practices in response to AI ?

Section 4 : Reflection on the process

How our contribution evolved during Phase 2

1. Starting point : a position judged too neutral

Our initial Phase 1 submission approached the question from a student-centred angle : “How can AI be used to enhance learning without weakening students’ cognitive abilities ? ” The analysis identified real risks cognitive dependency, assessment reliability, socio-economic inequalities but remained deliberately balanced, presenting a restrictive and an integrative perspective before settling on a cautious middle ground. The proposed contribution was a set of pedagogical recommendations and a policy reference framework.

Feedback from our project tutor, Prof. Stéphane Frénot, was direct and valuable. His written assessment noted that our position was :

« Un peu neutre/naïve » — « vous ne vous engagez pas trop ni dans un sens ni dans l’autre »

(“somewhat neutral and naive” — “you do not commit strongly in either direction”)

This critique was decisive on two levels. First, it pushed us to reframe the central question entirely shifting from how students use AI to what the institution itself must structurally become. Second, it pushed us to take a real argumentative stance rather than mapping perspectives from a safe distance. The result was a fully rewritten submission built around a single structural thesis : the student using AI to meet institutional expectations is not the anomaly the institution that has not updated its model is.

2. The decisive provocation : the AI tutor of 2130

The most transformative input came from a specific provocation in Frénot’s feedback. He invited us to reason as follows :

« Supposons que vous inventiez l’IAg de 2130, qui est spécialisée dans la pédagogie. Supposons qu’elle soit parfaite. Qu’elle est alors la différence par rapport à un enseignant ? »

(“Suppose you were to invent the generative AI of 2130, specialised in pedagogy. Suppose it were perfect. What would then be the difference compared to a human teacher ?”)

This thought experiment became the backbone of our proposal. Rather than asking “how should institutions adapt to AI ?”, we reframed the question as : “what is the irreplaceable value of a human institution once AI tutoring is perfect ?”. This shift led directly to our Position C — the long-horizon institutional design frame and anchored our three-axis proposal.

Frénot also introduced a distinction between two modes of AI use that influenced how we thought about student-AI interaction : AI as an occasional external resource (like a search engine), and AI as a continuous integrated companion (like glasses). This nuance shaped our thinking on cognitive dependency, evaluation design, and the conditions under which human guidance remains essential.

3. Influence of the keynotes

Five keynote sessions during Phase 2 contributed meaningfully to the evolution of our thinking :

• Keynote 1 — Caroline Beslin

  • Emphasis on social relevance, implementation clarity, and emotional engagement in communication

  • Led us to reframe our proposal as a systemic transformation, not merely a technical or regulatory recommendation

• Keynote 2 — Alexandra Pregent/ AIOLIA Project (AI Ethics & Governance)

  • Contributions on harmful AI uses, surveillance, misinformation, and epistemic gaps

  • Reinforced our argument that governance cannot be static it must be an ongoing institutional process

  • Helped us integrate ethical dimensions into our structural axes

• Keynotes 3 & 4 — Communication, implementation, and human learning limits

  • Questions on emotional dependency and cognitive atrophy from AI use

  • Reinforced our position that human interaction and institutional structures remain indispensable

  • Strengthened Axis 2 (longitudinal, process-based assessment) as a response to AI’s growing ability to simulate outputs

• Keynote 5 — Human-centered AI

  • Examples of biased AI systems (hiring tools, image classification failures) reinforced the need for governance and ethical oversight

  • Confirmed that institutional adaptation cannot be reduced to technical efficiency

4. Platform discussions and cross-project learning

Engagement with other participants on the challenge platform was a significant source of refinement :

• Gaëlle Méchin raised the SLAM model (JGU 5)

  • This pushed us to clarify our position on single-moment vs. longitudinal evaluation

  • We acknowledged SLAM’s strength in real-time oral assessment while arguing that it still functions as a snapshot, and that combining it with longitudinal tracking would be more robust

• Ndeye Sophie Seck asked how longitudinal reasoning traces can remain authentically human

  • This sharpened our thinking on continuous, multi-modal, supervised assessment as the only reliable alternative to AI-generatable outputs

• Guilhem Poutrain-Mari challenged the institutional feasibility of our proposal

  • Questions on accreditation cycles and teacher workload led us to emphasize gradual transformation, adaptive governance, and building on existing initiatives rather than radical overhaul

• Dr Christelle Scharff raised the comparison between MOOCs and AI tutors

  • This led us to distinguish between passive information access (MOOCs) and interactive, immediate AI assistance

  • We clarified that AI’s value as a tutoring tool depends critically on epistemic quality, guidance structures, and the continued role of educators as pedagogical mentors

What changes we made as a result

The feedback and discussions of Phase 2 produced concrete changes to our contribution :

• We moved from a broad diagnostic overview to a committed structural proposal with three clearly articulated axes

• We adopted the long-horizon frame (Position C) as our primary analytical lens, rather than presenting all three positions as equally valid

• We reinforced the governance axis (Axis 3) by specifying that governance bodies must have binding decision-making power, not merely advisory roles

• We integrated the distinction between AI assistance and institutional transformation more explicitly throughout our argumentation

• We introduced nuance on implementation conditions, acknowledging accreditation constraints and advocating for gradual, experimentation-based transformation

• We clarified the difference between AI access to information and genuine pedagogical guidance, a distinction sharpened by Dr Scharff’s questions

How this process strengthened our final proposal

The Phase 2 process transformed what began as a relatively descriptive critique into a genuinely structural, forward-looking, and implementable proposal.

The tutor’s provocation gave us our core analytical frame. The keynotes gave us breadth across governance, ethics, communication, and human cognition. The platform discussions gave us precision : they forced us to test our arguments against real objections around feasibility, assessment authenticity, and cross-disciplinary applicability.

Rather than proposing a sudden institutional rupture, our final contribution advocates for a direction of travel anchored in the long-horizon question of what human institutions uniquely offer with concrete structural moves that can begin now and remain coherent with where higher education needs to go.

The process itself modeled something our proposal argues for : genuine learning requires iteration, challenge, and guided reflection. It does not emerge from isolated output production.

Section 5 : Final contribution

Our final contribution document (including full proposal and pitch video) is available via the link :

Proposal

Pitch