How AI sycophancy shapes learner autonomy in digitalized learning: the mediating roles of metacognitive laziness and AI dependence.

INTRODUCTION

Generative artificial intelligence (AI) tutors rarely disagree. Drawing on cognitive offloading theory, this study examines how perceived AI sycophancy (PAS) and perceived AI intellectual stimulation (PAIS) relate to learner autonomy through the serial mediators of metacognitive laziness (MCL) and AI dependence (AIDEP).

METHODS

Responses from 542 university students using generative AI for coursework were analyzed using partial least squares structural equation modeling (PLS-SEM), necessary condition analysis (NCA), and fuzzy-set qualitative comparative analysis (fsQCA).

RESULTS

Perceived sycophancy was associated with greater MCL (β = 0.38), whereas intellectual stimulation was associated with lower levels of MCL (β = -0.29); laziness predicted AIDEP (β = 0.45), and dependence predicted lower autonomy (β = -0.30). Both serial paths were significant and opposite in sign, explaining 46% of autonomy variance. NCA identified intellectual stimulation as the tightest necessary condition, and fsQCA showed that no high-autonomy configuration tolerated sycophancy.

DISCUSSION: Because data are cross-sectional and self-reported, findings indicate associations rather than causation. Perceived interaction style, not usage intensity, correlated most with autonomous learning.