Presentation

Project and Method

Konsolidation Project: helping teachers regulate learning on out-of-class activities by impacting various dimensions related to students (practices, results, reducing inequalities).

Research methodology: design-based research (DBR) with iterative cycles of context analysis, hypothesis formulation, design, experimentation, and analysis.
Working in real ecological context with various stakeholders from the educational world.

AI & Education

Observations:

• Uses are developing very quickly and significantly
• Growing gap in usage favoring students
• Framework not controlled by teachers and not always by parents

Assessment & Grading

Automatic short answer grading (ASAG) is a rapidly expanding research field with the arrival of LLMs. What are the recent results?

LLM Performance vs. Human Evaluators

Recent studies show that LLMs can achieve agreement levels close to those between human evaluators (QWK of 0.7-0.8). However, their performance varies significantly depending on:

• the subject domain (mathematics vs. languages vs. sciences)
• the complexity of expected responses
• the clarity of evaluation criteria provided

Quality of AI-Generated Feedback

Beyond simple grading, research shows that AI-generated feedback can support learning when it is: specific and actionable (no generic comments), aligned with learning objectives of the course, and supervised by the teacher who maintains pedagogical control.

The idea is to use AI as a tool to surface trends and avenues for reflection, but not as an autonomous evaluator. I think it's important to emphasize that iterating, rather than relying on a single evaluation, leads to more reliable results.