Not All Cognitive Offloading Is Equal: Distinguishing Dependent and Autonomous Offloading to Generative AI
Citation. Zhu, Q., Li, X., Dong, Y., Chang, P., & Fan, M. (2026). Not all cognitive offloading is equal: Distinguishing dependent and autonomous offloading to generative AI. Frontiers in Psychology, 17, 1878629. https://doi.org/10.3389/fpsyg.2026.1878629
Summary. Zhu et al. distinguish dependent cognitive offloading—delegating core thinking to AI—from autonomous offloading, in which AI output scaffolds but does not replace the user’s thinking. In a three-wave survey of 589 Chinese university students and early-career knowledge workers, the modes had comparable perceived immediate benefits but different associations with later perceived cognitive outcomes. Dependent offloading was associated with cognitive-agency transfer and lower intrinsic motivation, whereas autonomous offloading was associated with higher intrinsic motivation and more favorable perceived capability, creativity, deep processing, and independent judgment.
Key claims
- Dependent and autonomous cognitive offloading were empirically distinct in the authors’ 11-factor CFA and only weakly correlated (r = 0.08), so frequency of AI use does not by itself characterize the mode of cognitive engagement.
- Dependent offloading at Time 1 was associated with more cognitive-agency transfer at Time 2 (β = 0.35) and lower intrinsic motivation (β = -0.23); autonomous offloading was not significantly associated with agency transfer (β = -0.05, p = 0.18) but was positively associated with intrinsic motivation (β = 0.28).
- Cognitive-agency transfer was negatively associated with all four perceived Time-3 outcomes (βs = -0.15 to -0.29), while intrinsic motivation was positively associated with them (βs = 0.26 to 0.36); bootstrapped indirect associations supported both pathways for dependent offloading and the motivational pathway for autonomous offloading.
- Metacognitive monitoring attenuated, but did not eliminate, the association between dependent offloading and cognitive-agency transfer (β = 0.48 at low monitoring; β = 0.21 at high monitoring), and it did not significantly buffer the negative association with intrinsic motivation.
- Both modes had the same correlation with perceived immediate benefit (r = 0.22), while immediate benefit had weak or non-significant associations with later perceived cognitive outcomes; short-run utility therefore did not distinguish the modes in this survey.
Evidence & limitations
- Design: three-wave, two-week-lagged survey; 589 retained participants (54.3% undergraduates, 25.3% master’s students, 8.0% doctoral students, 12.4% young professionals), with newly developed four-item AI-specific measures and covariate-adjusted SEM.
- The design supports temporally ordered associations, not causal mediation or demonstrated change in cognitive ability: all focal downstream outcomes were self-reported, baseline mediator/outcome measures were absent, and the authors’ indirect effects remain correlational.
- The measures show internal consistency and within-study discriminant evidence but have not yet been tested against adjacent instruments such as AI dependency, automation complacency, blind trust, critical AI literacy, or self-regulated learning; the primarily Chinese student sample and small worker subsample limit workplace generalization.
- Public provenance is verified: the Frontiers version-of-record PDF at the DOI has the same 16-page content and exact normalized extracted text as the ingested Drive file, including the same binary SHA-256 hash.