When Peers Outperform Ai (And When They Don'T): Interaction Quality Over Modality

Date

01/2026

AI Overview

Why it matters

As AI assistants increasingly substitute for peer interactions in classrooms, fundamental questions arise about how algorithmic partners change social learning mechanisms like shared regulation, explanation, and curiosity.

Study at a glance

Question: How do student interactions and learning experiences differ when collaborating with peers versus an AI assistant?
Setting and data: 36 undergraduate students at a large research university, interacting remotely over Zoom and Miro.
Method: Randomized lab experiment with computational and qualitative discourse analysis and post-session surveys.
AI tools: Claude 3.5 Sonnet

Key findings

High-quality peer interactions generated greater curiosity and engagement than AI, but low-quality peer interactions scored worse than AI on learning confidence and topic interest.
AI learners reported high confidence in explaining concepts but lower curiosity and new ideas, frequently treating the assistant as a transactional answer source rather than a collaborator.
Interaction quality was predicted by early dialogue behaviors—such as question-asking, turn balance, and expressing social curiosity—rather than by students' individual personality traits.

Summary drafted with AI from the full paper and may contain errors. See the full paper for details.

As AI increasingly enters the classroom, what changes when students collaborate with algorithms instead of peers? We analyzed 36 undergraduate students learning graph theory through peer collaboration (n=24) or AI assistance (n=12), using discourse analysis to identify interaction patterns shaping learning outcomes. Results reveal a collaboration quality divide: high-quality peer interactions generated curiosity and engagement that AI couldn't match, yet low-quality peer interactions performed worse than AI across dimensions. AI showed a paradoxical pattern, building confidence in knowledge while reducing curiosity and deeper engagement. Interaction quality emerged from dynamic patterns rather than individual traits, with early discourse markers predicting outcomes. Students treated AI as a transactional information source despite its collaborative design, revealing fundamental differences in human versus algorithmic engagement. Our findings suggest AI in education need not replace peer learning but can recognize struggle and support both peer and AI interactions toward productive learning experiences.