Learning with a Scripted AI Tutor or a Virtual AI Learner: Experimental Investigation of Collaborative Learning Using CoCot
Yugo Hayashi, Shigen Shimojo, Tatsuyuki Kawamura · Computer-supported collaborative learning/The Computer-Supported Collaborative Learning Conference · 2024
This study investigated effective methods for using conversational agents for learnerlearner support in a collaborative explanation task using concept maps.Scripted facilitation methods that use conversational agents were compared with the proposed method that integrates the reciprocal teaching strategy.This method uses an ACT-R cognitive architecture-based virtual peer learner agent to identify, reason, and construct a concept map from dyads in real time.An experiment was conducted under two conditions: 1) a tutor as a mediating agent that prescribes metacognitive scripts, and 2) an adaptive virtual peer agent that prescribes concept maps.The results demonstrated that both methods improved the learning performance compared with the control condition.Notably, learners working with the peer learner agent made better use of procedural knowledge during tasks.This study suggests that a "horizontal" learning approach with the agent as a peer is as effective as the methods with facilitated instruction using scripts.