A Hybrid “Probabilistic Scaffold” Model for Enhancing Reading Literacy, Numeracy, and AI Literacy among Indonesian Primary School Students

This design study develops a structured pedagogical framework positioning generative AI as a verification-dependent learning support for Indonesian 4th-5th grade students. The framework implements a sequential process (Prompt → Verify → Diagnose → Revise → Reflect → Transfer) bounded by ethical guidelines and local cultural contexts. The research addresses student bypassing of initial reasoning, absence of age-appropriate AI literacy assessment, digital access disparities, insufficient verification practices, and potential erosion of academic integrity. Using a quasi-experimental design (n≈200 across 4 schools; 12 weeks), the study compares an AI-integrated curriculum with conventional instruction, controlling for pre-test performance, socioeconomic indicators, and gender. The intervention balances AI-mediated learning (25-35%) with analog activities (65-75%) to preserve direct cognitive engagement. Assessment instruments include reading comprehension measures, contextual mathematics problems, a four-dimensional AI literacy rubric, interaction logs, classroom observations, motivation scales, and verification records. Anticipated outcomes include normalized learning gains of 0.35-0.45 in reading and 0.30-0.40 in mathematics; advancement to Level 3 AI literacy; reduction in AI dependence from 60% to 35-40%; increased higher-order questioning; and equitable outcomes across settings through offline resource distribution. The theoretical framework proposes that prompt quality influences learning outcomes through verification practices, metacognitive awareness, and ethical understanding. Implementation challenges include novelty effects, variable teacher adherence, technology access inconsistencies, and ethical standards application. The research aims to shift elementary AI education from answer retrieval toward evidence-based inquiry with ethical awareness.