Understanding Student Dependency on AI: The Role of AI Literacy, Academic Self-Efficacy, and Resource Management Strategies

Synthesis: Among 478 Israeli higher-education students, AI literacy is a double-edged factor: its skill-based dimensions (using/understanding AI) were positively associated with reported AI dependency, while self-efficacy (both academic and AI-specific) and effort AI Regulation in Education were negatively associated. AI literacy alone does not protect against overreliance — it can even enable it. Self-efficacy beliefs and self-regulated learning practices appear at least as important as technical skill for balanced, intentional AI engagement.

Key Findings

• Multidimensional AI literacy operates in opposing directions. The skill-based dimensions (using/understanding AI: β = 0.404, the strongest predictor; detecting AI: β = 0.108) were positively associated with dependency. In contrast, AI self-efficacy (confidence in one's own AI competence) was negatively associated (β = −0.132). Academic self-efficacy correlated negatively with dependency on its own (r = −0.24) but entered the full model positively (β = 0.20). The pattern suggests a compensatory mechanism for AI-specific confidence: strong skills enable offloading, while strong AI self-efficacy buffers against it (Cognitive Offloading, Self-Efficacy).
• The model explained 27.3% of variance in AI dependency (F(8, 469) = 22.03, p < 0.001; R² = 0.273, adjusted R² = 0.261; N = 478), which the authors read as support for treating AI literacy as multidimensional rather than as a single aggregate score.
• Effort regulation predicted lower dependency (β = −0.13); general academic help seeking predicted higher dependency (β = 0.12); time/study management was not significant (β = 0.05, p = 0.374). The help-seeking measure captures human sources, so the finding does not show AI substituting for human help.
• Four student profiles emerged (K-means cluster analysis): HL (high literacy, low dependency — most adaptive), MM (moderate/moderate), LL (low/low), and HH (high literacy, high dependency). The HL profile scored highest on academic self-efficacy, time/study management, and effort regulation; HH — despite strong technical knowledge — did not lead on any self-regulation measure, suggesting reliance may undermine learner autonomy and create an illusion of competence.
• The dual structure within AI literacy (skills that enable use vs. beliefs that govern reliance) is a novel contribution to the AI-dependency literature, supporting calls to treat AI Literacy as multidimensional rather than unitary.

The compensatory self-efficacy mechanism

The central insight is that skill and confidence pull in opposite directions in predicting dependency. AI literacy curricula focused narrowly on technical skill (how to prompt, how to use tools) may inadvertently increase dependency. By contrast, curricula that also build students' sense that they can "think with AI rather than through AI" — and their general academic confidence — may protect against overreliance. The authors frame this as a shift from a technological-instrumental conception of AI literacy toward a pedagogical, behavioral, and emotional one foregrounding self-regulation, self-awareness, and learner responsibility.

What this means for practice

• Learners. Treat AI-literacy training as necessary but not sufficient: the skill-based dimensions of AI Literacy (using/understanding AI and detecting AI) were positively associated with dependency, so pair tool skill with deliberate judgment about when to hand work over.
• Learners. Build the counterweight yourself — academic and AI Self-Efficacy and effort AI Regulation in Education were all negatively associated with dependency, so keep practicing work you can verify without AI rather than letting confidence rest on the tool.
• Learners. Scaffold regulation, not just prompting: document your process, reflect on and critique AI outputs, and manage time and effort deliberately, because the more dependent students reported weaker effort regulation and time management.
• Instructors. Differentiate support by profile: the study's cluster analysis identified four learner profiles that combine literacy and dependency differently, so give foundational AI skills to novices and critical, reflective-use training to those already skilled.

Limitations

• Cross-sectional, self-report online survey of 478 students at Israeli higher-education institutions recruited through a commercial panel (Panel4All); the design supports associations, not the causal reading of the proposed "compensatory" mechanism.
• Every construct was measured with self-report Likert items, and the ethics and emotional-regulation AI Literacy subscale had internal consistency of α = 0.64, below the conventional 0.70 threshold.
• The sample is confined to a single country (Israel) and one data-collection window (August–September 2025), so the profiles may not transfer to other national or institutional contexts.

Citation

Maizel, H., Kalman Halevi, M., Sarid, M., & Tutian, R. (2026). Understanding Student Dependency on AI: The Role of AI Literacy, Academic Self-Efficacy, and Resource Management Strategies. Education Sciences, 16(7), 1123.