Does school-based AI education narrow readiness gaps? The role of prior agency-related learning

Synthesis: Liang, Yau, Meng, Chiu, King, Yam & Chai (2026) followed 752 Hong Kong junior-secondary students across a year of school-based AI instruction and found that structured curricula narrowed psychological AI-readiness gaps (confidence, motivation, ethical awareness) but not cognitive ones — objective AI-literacy gaps between high-agency learners and their peers persisted. Guided by Social Cognitive Theory, prior self-initiated AI learning predicted who benefited most, so school AI education acts as a psychological catalyst but not a full cognitive equalizer.

The study

• Sample: 752 Hong Kong junior-secondary students (Mage = 14.06), 2024–25 school year, across three prior-learning profiles:
  • High-agency learners (N = 183) — one year of self-initiated AI learning (e.g., online tutorials) only
  • Curriculum-supported learners (N = 369) — one year of school AI instruction only
  • Low-agency late learners (N = 200) — no prior AI learning of any kind
• Design: two-wave multigroup latent change score (LCS) model, before/after a year-long school AI curriculum (12 chapters across AI foundations, ML & applications, AI for social good, ethics/responsible use)
• Outcomes: psychological AI readiness (confidence, perceived AI knowledge, social-good orientation, ethical awareness, learning intention) plus an objective 25-item AI-literacy test (cognitive readiness)

Baseline: prior agency predicts readiness

• High-agency learners entered with the strongest psychological and cognitive readiness; curriculum-supported learners sat in the middle; low-agency late learners scored lowest
• Self-initiated AI learning functioned as a behavioral pathway to readiness, while prior school instruction was an environmental affordance that only partially substituted for it

After one year of instruction: psychological gains, cognitive persistence

• All three groups improved on both psychological and cognitive readiness after the year-long curriculum
• Psychological gaps narrowed substantially — low-agency late learners showed the largest gains and approached parity with curriculum-supported learners, evidence that structured instruction can be a psychological catalyst / compensatory environment
• Cognitive gaps persisted — group differences in AI-literacy change were non-significant, and high-agency learners still outperformed peers at post-test; a Matthew-effect pattern where school AI education "raised the floor but did not fully level the playing field"

Why psychological and cognitive readiness diverge

• Psychological readiness is more immediately responsive to structured instruction (self-report of confidence, motivation, ethical orientation)
• Objective AI literacy requires cumulative exposure and sustained self-initiated exploration — perceived gains do not automatically translate into objective understanding
• The authors argue school AI education works synergistically with prior agency-related learning rather than replacing it: curricula provide access/guidance, while self-initiated engagement helps learners interpret and extend those affordances

What this means for practice

• Instructors. Pair structured AI curricula with inquiry-based projects, reflective discussion, and authentic tasks that cultivate self-initiated engagement rather than content delivery alone — prior self-initiated AI learning, not school instruction, predicted the largest cognitive gains.
• Administrators. Treat equity as more than curriculum access: ensure early and timely instruction, teacher preparation, and out-of-school enrichment so agency-related learning becomes a shared capacity rather than a stratified advantage.
• Policymakers. Do not read improved confidence, motivation, and ethical awareness as evidence that AI-literacy gaps have closed — psychological gaps narrowed substantially while objective AI-literacy differences persisted.
• Learners. Treat the school curriculum as an entry point rather than a ceiling: high-agency learners still outperformed peers at post-test after a year of instruction, so pursue self-initiated exploration and projects beyond it.
• Researchers. Evaluate single-cohort gains against an objective AI-literacy measure and compare growth across learner subgroups, because the psychological improvements in this study were concentrated in self-report and approached parity while between-group cognitive change stayed non-significant.

Limitations

• Only two measurement waves, with outcomes captured immediately after one year of instruction: the study cannot show whether psychological gains persist, whether cognitive gains consolidate, or whether the readiness gaps widen across later school years.
• Agency-related learning was operationalized categorically as three prior-learning profiles (high-agency N = 183, curriculum-supported N = 369, low-agency N = 200) rather than as a continuous construct, so the design could not capture change in agency itself.
• Outcomes relied primarily on self-report indicators plus a 25-item AI-literacy test, which may not represent students' authentic engagement or practical competence.
• Participants were 752 junior-secondary students (Mage = 14.06) in Hong Kong, so both the curriculum structure and the readiness patterns may differ in other systems.

Citation

Liang, S., Yau, K. W., Meng, H., Chiu, T. K. F., King, I., Yam, Y., & Chai, C. S. (2026). Does school-based AI education narrow readiness gaps? The role of prior agency-related learning. Computers & Education, 256, 105745.