Impact of AI Support with Different Level of Learner Agency on Behavioral Engagement in Providing and Uptaking Peer Feedback

Abstract

This study examined how AI support with varying levels of
learner agency influences behavioral engagement during peer
feedback construction and uptake. Ninety university
students in 30 triads were randomly assigned to three
conditions: Control (no AI support), AI-directed (AI
provides finished products), or AI-assisted (AI provides
guidance and prompts). A conversation analysis framework
was proposed for examining learners’ effort in human-AI
conversations: elaboration level, topic pursuit, difficulty
response, and repair pursuit intensity. Results revealed
task-dependent patterns: participants showed greater
response depth and topic extension during feedback
construction, but more clarification-seeking during
feedback uptake. Significant interaction effects emerged
between AI support type and learning task. AI-assisted
learners displayed adaptive engagement—more elaborate
responses when constructing feedback, shifting toward
verification behaviors during uptake. AI-directed learners
maintained consistent engagement across tasks. These
findings suggest that AI in assistant role preserves
learner agency fosters task-adaptive engagement, while AI
in directive role produces undifferentiated engagement
patterns.