Use unlearning-based simulated students as scalable practice environments for teacher professional development¶

Strategy · All strategies
Evidence · no claims cited

Description¶

The article's implications section proposes that because the unlearning approach can generate simulated students at different knowledge levels, it can populate training environments where teachers or tutors rehearse pedagogical skills. The authors state "it could support training environments in which teachers or tutors practice responding to varied learner needs", offering a wider range of learner profiles than role-play or live coaching typically allows. This extends the teachable-agent pipeline from student simulation to educator practice infrastructure.

Design Implications¶

Context¶

Requirements¶

• The unlearning pipeline must be able to generate simulated students at multiple knowledge levels, as shown in Fig. 4

Constraints¶

• Educational applicability has not yet been tested in human studies; the authors call for human studies examining whether the method supports improved learning outcomes

Target Learners¶

• Teachers and tutors in professional development

Target Learning Goals¶

• Rehearsing pedagogical skills such as responding to varied learner needs and delivering corrective feedback

Affordances¶

• Unlearning Based Teachable Agent Pipeline

Related Strategies¶

• Use synthetic LLM-generated error datasets for AI tutors, teachable agents, and teacher professional development
• Use simulated student sections as a scalable supplement to teacher training and lesson preparation
• Extend simulated teaching practice to student assessment and teacher screening uses

Examples¶

-

Key Sources¶

• Jiajia Song, Zhihan Guo, and Jionghao Lin. (2026). Simulating Novice Students Using Machine Unlearning and Relearning in Large Language Models. https://arxiv.org/abs/2603.26142