AI in education

EduGenie TA ČR

Effective and Safe Use of Generative Artificial Intelligence Tools in the Student Learning Process

The project looks for the conditions under which pupils can use generative AI tools in class effectively and safely. It pays particular attention to differences between groups of pupils: gender, socio-economic background and special educational needs. A mixed-methods research design evaluates how AI affects the learning process.

Based on the findings, the consortium will develop, pilot and provide to schools an interactive LLM-based teaching tool that adapts to individual pupils' needs. The tool should deepen understanding of the curriculum, encourage critical thinking and creativity, and help pupils acquire new knowledge and skills.

Research questions

• When does an AI tutor support learning, and when does it replace the effort learning depends on?
• Do the benefits and risks differ across groups of pupils, and could AI widen existing gaps rather than close them?
• What does a safe AI tool for children look like in practice?

Where agents come in

A good tutor is a sequential decision-maker. It chooses what to ask next, when to give a hint and when to hold back, and it adapts to a pupil it gets to know over time. This is where our research on LLM agents meets education:

• Memory: a tutor that remembers a pupil's progress and misconceptions across sessions;
• In-context learning: a tutor that adapts its teaching strategy from the pupil's responses;
• Safety: guardrails suitable for children, and a tutor that supports thinking instead of handing out answers.