AI Literacy Is Not a Switch: Understanding the Staged Construct
The most common misconception in AI training programs: AI literacy is binary — you either have it or not. Research shows the opposite. AI literacy is a staged developmental construct moving along three axes.
1. Five convergent stage models (Guide 1, §2.1)
| Model | Stages | Setting | Evidence |
|---|---|---|---|
| DALF (Developmental AI Literacy Framework) | 3: foundational awareness → reflective application → integrated professionalization | Counselor education | (E) conceptual |
| AI Fluency Framework | 6: exposure → … → strategic, critical, ethical, contextual | Higher education | (E) conceptual |
| RAIL-Ed (Responsible AI Literacy in Education) | integrative, developmental, dialectical; ethics as constitutive | Teacher education | (E) synthesis of 67 studies |
| Five-stage continuum (transfer corpus) | Avoidance ↔ uncritical reliance (extremes) | Higher education | (E) conceptual |
| AI Mindset (triadic) | dispositional × contextual × technological | Educators | (C) empirically tested |
2. Three axes of the construct
- Competence depth (know → use → evaluate → create/integrate) — evidenced by the psychometric A-factor (one dominant latent factor across interaction tasks, 44% variance; N=517).
- Autonomy (dependence → independence) — from DALF's parallel to professional development.
- Disposition (attitudes, self-efficacy, mindset) — from AI Mindset's empirically tested triadic structure.
3. What AI literacy is NOT (Guide 1, §2.2)
- Not self-assessment. Self-reported and objective measures correlate only weakly.
- Not tool proficiency. Definitions emphasizing programming/statistics miss what workers actually need: interpreting outputs, selecting tools, ethical judgment.
- Not a one-shot workshop. The LLM-fallacy mechanism shows users systematically attribute LLM outputs to their own competence.
4. Implication for curriculum design
Tier the curriculum to the continuum (Awareness → Application → Evaluation → Creation) and map each stage to concrete job tasks. See Role-specific task matrices.
Further reading
Sources: Guide 1 (Implementation), §2.1/§2.2; corpus of 16,459 papers (as of 2026-10). EOF echo "en2 done"