Home | Publications | OBB+26

Should AI Ask First? Investigating the Effects of Proactive vs Reactive AI Mentoring in Self-Directed Learning

MCML Authors

Link to Profile Enkelejda Kasneci

Enkelejda Kasneci

Prof. Dr.

Core PI

Human-Centered Technologies for Learning

Abstract

Motivated by decades of research on tutoring and help-seeking in intelligent learning environments, we ask an important but underexplored question: Should an AI mentor wait for learners to ask for help, or ask first? We compare proactive and reactive initiative policies in a between-subjects experiment ((N=81) using a 14-min instructional video about Bayes’ theorem. The proactive mentor provided brief prompts on natural task boundaries using lightweight personalization and a cooldown mechanism to prevent overload. Proactive mentoring significantly increased interaction volume (10.5 vs. 2.9 messages on average, (p < 0.001), reduced no-interaction rate (8.1% vs. 31.8%, p = 0.013), and reduced off-topic messaging (2.0% vs. 6.8% of messages, p < 0.01), shifting communication towards help-accepting behaviors and sustained conversations. Cluster analysis revealed a distinct help-accepting profile (observed only in the proactive condition). Knowledge gains showed a non-significant upward trend for the reactive group. The learners in the reactive group reported greater perceived choice and control, highlighting a design trade-off between interaction and agency. Together, these findings identify initiative-coupled scaffolding policy as a key determinant of help-seeking and learner-AI interaction patterns in self-directed learning settings.

inproceedings OBB+26

AIED 2026

27th International Conference on AI in Education. Seoul, Republic of Korea, Jun 27-Jul 03, 2026.
Conference logo
A Conference

Authors

K. Otmani • A. Bodonhelyi • B. Bühler • E. Kasneci

Links

DOI GitHub

Research Area

 B3 | Multimodal Perception

BibTeXKey: OBB+26

#p-kasneci-enkelejda
Back to Top