Designing with memory: Multi-Agent LLM systems for adaptive lesson planning in AI literacy education

• In Event: AERA Poster Session Friday 1:45 pm
  In Poster Session: Developing AI Literacy and Teacher Professional Competence

Abstract

This study introduces a generative AI–enabled co-design tool that supports teachers in developing culturally responsive, AI-integrated lesson plans. The system features curriculum, ethics, and feedback agents, along with a semantic memory module that personalizes support across pedagogical and cultural contexts. Six middle school teachers engaged with the tool over three weeks in co-design workshops. Analysis revealed three user profiles—guided explorers, adaptive improvisers, and critical skeptics—each representing distinct approaches to collaborating with AI. Teachers progressively adapted AI outputs to align with their goals and classroom needs. Findings highlight the role of teacher agency and interpretability in designing human-centered, adaptive AI systems for education.

Author

• Xiaoxue Du, MIT Media Lab