Experimental Investigation on Reciprocal Teaching Using Misconception-Based Teachable Agents in Collaborative Learning

This study examines methods to promote mutual explanatory activities in collaborative learning using a teachable virtual peer learner.Its purpose is to determine whether the quality (error level) of knowledge generated by a teachable virtual peer learner can encourage learners to reevaluate their understanding.This study uses the ACT-R cognitive architecture to examine a teachable agent that learns from learners' concept maps and generates feedback maps.Further, a comparative experiment is conducted on a non-agent, an erroneous knowledge agent, and a non-erroneous agent.Results reveal that the use of a teachable agent enhances the learning activities of learners.Specifically, when learners interact with the erroneous knowledge agent, they actively revise their concept maps and, thereby, improve learning outcomes.This revision process was prompted by observing incorrect maps and encouraging learners to review and correct their knowledge.