L'IA pour mieux apprendre et apprehender L'IA

Frédéric Alexandre, Marie-Hélène Comte, Aurélie Lagarrigue, Thierry ViévilleSPIRE 2023 · OpenAlex

Abstract

National audience. This paper explores the dual role of Artificial Intelligence in education: as a sophisticated tool for personalizing human learning and as a critical subject of study for modern citizenship.

Introduction

As Artificial Intelligence (AI) becomes ubiquitous across all sectors of society, the field of education faces a unique paradox. On one hand, AI offers unprecedented opportunities for creating efficient, personalized learning pathways that can adapt to the unique needs of every student—a task that human teachers, often stretched thin, struggle to perform manually. On the other hand, there is a growing risk that both students and educators will become passive consumers of "black box" technologies, potentially eroding individual agency and critical thinking. This paper argues for a framework where AI is used to delegate "mechanizable" tasks, thereby freeing human intelligence for higher-level cognitive and creative goals.

Methodology: From Algorithms to "Unplugged" Activities

The authors propose a multi-faceted approach to integrating AI into pedagogy, synthesizing insights from digital sciences and cognitive neuroscience.

  1. Adaptive Algorithms: By analyzing "learning traces"—digital footprints such as mouse clicks, response times, and sensor data—AI systems can adjust the difficulty of exercises in real-time. This ensures that students remain engaged without becoming overwhelmed.
  2. Computational Educational Science: This emerging field uses AI models to represent functional brain processes, creating a scientific mirror for human cognition.
  3. "Unplugged" Computer Science: To demystify AI, the authors advocate for physical, non-digital activities. For instance, physical models can be used to teach complex concepts like reinforcement learning without the need for a screen, making the "mechanics" of AI tangible.
A physical model used to demonstrate reinforcement learning concepts without digital screens.

Case Study: The KidLearn Project

A primary example of these principles in action is the KidLearn project. This initiative utilizes specific pedagogical scenarios, such as a virtual merchant game where students practice buying objects or giving change, to track and support progress in arithmetic.

The KidLearn interface showing a merchant scenario where students practice arithmetic by giving change.

The system monitors student interactions to build a model of their current knowledge state, allowing it to recommend the next best activity. This data-driven approach ensures that the curriculum is always tailored to the learner's specific progress.

Key Insights and Results

The research highlights several transformative benefits of AI in the classroom:

  • Targeting the Zone of Proximal Development (ZPD): AI can algorithmically identify the "sweet spot" between an exercise being too easy (leading to boredom) and too difficult (leading to frustration), thereby maximizing learning efficiency.
  • The "Non-Judging" Machine: One of the most significant psychological benefits of AI tutors is their lack of judgment. Students often feel more comfortable making mistakes and experimenting when interacting with a machine, which can significantly boost engagement and reduce the fear of failure.
  • Teacher Empowerment: Rather than replacing teachers, well-implemented adaptive tools empower them. By automating rote content transmission, AI allows educators to focus on complex, project-based interventions and social-emotional support.
  • Citizen Literacy: The authors emphasize that understanding AI's "unpredictable" nature—where outcomes are driven by data parameters rather than hard-coded instructions—is vital for modern literacy. Initiatives like the "Intelligence artificielle avec intelligence" MOOC are crucial for training educators to navigate this landscape.

Limitations and Future Outlook

Despite the potential, the integration of AI into education is not without risks. The authors acknowledge significant ethical concerns regarding data privacy and the "omnipresent" nature of student tracking. There is a real danger of reinforcing social inequalities, often referred to as the "digital divide" or illectronisme.

Furthermore, the design of pedagogical tools must be carefully managed. Overly "gamified" interfaces can increase cognitive load, causing students to focus on game mechanics rather than learning objectives. As the field of "computational educational science" matures, future research must bridge the gap between laboratory-based neuroscience and the complex, messy reality of the classroom. Ultimately, the goal remains a "human-in-the-loop" system where AI serves to enhance, not replace, the human element of teaching.

Summary updated

Abstract

National audience
BibTeX
@inproceedings{alexandre2023l,
  author = {Frédéric Alexandre and Marie-Hélène Comte and Aurélie Lagarrigue and Thierry Viéville},
  title = {{L'IA pour mieux apprendre et apprehender L'IA}},
  booktitle = {SPIRE},
  year = {2023},
}
Figure