The Global Learning Crisis: 70% of 10-Year-Olds Cannot Read a Simple Text — AI Tutoring Closing Gaps at Scale
Objective
Assess the scope of global learning poverty, its structural causes, and evaluate the evidence base for AI-assisted personalized tutoring as a scalable intervention in low- and middle-income countries.
Methodology
Systematic literature review of 87 randomized controlled trials and quasi-experimental studies on EdTech and AI tutoring interventions published 2018-2024. Stratified by country income level, urban/rural setting, age group, and subject area. Meta-regression on learning outcome effect sizes. Case study analysis of 12 large-scale national deployments (India, Kenya, Peru, Indonesia, Nigeria, Ethiopia).
Findings
Learning poverty affects 70% of children in low- and middle-income countries (World Bank 2024). Root causes are structural: chronic teacher shortages (average pupil-teacher ratio of 52:1 in Sub-Saharan Africa), low instructional quality, poor attendance driven by economic pressures, and language barriers where instruction is in a non-native language for 40% of students.
31 SD on learning outcomes across 34 RCTs in LMICs — comparable to reducing class size by 10 students. 51 SD), suggesting strong equity potential. Cost per learning-adjusted year gained: $18-45 for AI tutoring vs $1,200+ for teacher training programs.
Offline-capable AI tutoring apps (functioning without reliable internet) have demonstrated sustained gains in Kenya, India, and Nigeria at scale. Key bottleneck: not technology availability but teacher adoption, device access, and electricity reliability.
Key Assumptions
- •Effect sizes from RCTs transfer partially to at-scale rollouts (estimated 30-40% attenuation typical for education interventions)
- •Literacy measures are comparable across country assessment frameworks
Limitations
- •Publication bias likely inflates average effect sizes — unpublished null results underrepresented
- •Long-term retention effects beyond 12 months understudied
- •Gender-disaggregated data missing from 40% of studies reviewed