When Infrastructure Meets Aspiration: AI's Contested Role in the Global Learning Crisis

Introduction

The 2030 deadline for Sustainable Development Goal 4 arrives in five years, yet the global education community remains trapped in a fundamental contradiction. We measure what's easy to count—enrollment rates, completion percentages, gender parity indices—while the global education crisis unfolds in plain sight within classrooms themselves. In 18 of 31 recently assessed low and lower-middle income countries, fewer than 10% of children achieve minimum proficiency in reading or mathematics by the end of primary school. This isn't a marginal implementation challenge. This is wholesale system failure.

The political economy of SDG4 target-setting reveals this evasion clearly. Targets 4.1, 4.2, 4.3, and 4.5 emphasize access, equity, and infrastructure—the visible, fundable elements that generate ribbon-cutting ceremonies and donor reports. What they systematically avoid are the three foundational challenges that actually determine whether children learn: instructional quality, early cognitive development, and teaching capacity. These aren't listed because they're politically inconvenient or considered intractable. They require decade-long commitments, resist quick measurement, and demand sustained domestic financing rather than project-based aid.

The data crystallizes this avoidance. Sub-Saharan Africa needs to triple its pre-primary teacher workforce and increase primary teachers by 50% to meet even basic access targets. Global estimates suggest 44 million additional teachers are required by 2030. Yet current SDG4 monitoring barely tracks instructional time, pedagogical practice, or whether teachers themselves possess the subject knowledge they're meant to transmit. We've constructed an accountability architecture that measures theater attendance while remaining studiously incurious about whether anyone can hear the actors.

The Infrastructure Reality Check

Into this measurement vacuum comes the perennial techno-optimism that characterizes education development discourse. Artificial intelligence, we're assured, can augment teacher capacity, personalize instruction, and leapfrog institutional constraints. The vision is seductive: every teacher and student equipped with an AI assistant, transforming pedagogical possibilities even in resource-constrained environments.

But examine the preconditions this vision requires. We find that one in four primary schools globally lacks electricity. Fewer than half have computers or internet for pedagogical use. In early 2021, under 10% of low-income country schools had adequate soap, water, and hygiene provisions. The infrastructure gap isn't ancillary—it's definitional.

Consider what "AI-assisted instruction" actually demands in practice. Reliable electricity for device charging. Internet connectivity for cloud-based systems or regular content updates. Devices themselves—tablets, smartphones, or computers. Technical support when systems fail. Digital content in local languages. And critically, teachers with sufficient digital literacy to operate these systems while managing classes that often exceed 60 students.

Some will argue for offline AI solutions, pre-loaded content, or solar-powered implementations. These exist. But they don't address the core constraint: teacher capacity. A teacher who struggles to teach reading using books and chalkboards won't magically become effective because an algorithm generates personalized worksheets. The pedagogy bottleneck precedes the technology question.

Where AI Might Actually Help—And Where It Won't

This doesn't mean AI has no role. But we must be precise about what problems it could plausibly address versus what problems require fundamentally different interventions.

AI could potentially help with administrative burden reduction. Teachers in LIC/LMIC contexts spend enormous time on attendance tracking, grade recording, and report generation—tasks that consume hours that could otherwise go to instruction or preparation. Simple AI tools (many requiring minimal connectivity) could automate these functions, returning 3-5 hours weekly to teachers. This is unglamorous but meaningful. If nothing else gets done, this is at least a positive step.

AI could assist with lesson planning for multi-grade classrooms. In rural areas where one teacher handles three grade levels simultaneously, AI systems could help sequence activities, suggest differentiated tasks, and provide age-appropriate materials. This addresses a real pedagogical challenge that human teacher training struggles to solve at scale.

AI-driven adaptive practice could work in specific contexts. Where students have any device access (even shared), algorithms can provide individualized mathematics or literacy practice calibrated to current ability. This doesn't replace instruction, but it could extend practice opportunities beyond what a single teacher can provide to 60+ students.

However, AI cannot address the fundamental problems. It cannot create the 44 million teachers needed globally. It cannot substitute for early childhood cognitive development that shapes life-time, learning trajectories before primary school begins. It cannot overcome the fact that many teachers lack subject matter knowledge themselves—no AI assistant makes an innumerate teacher effective at teaching mathematics.

Most critically, AI cannot solve the financing gap. The $97 billion annual shortfall represents the cost of adequate teacher salaries, functional school buildings, learning materials, and basic infrastructure. Some LIMC already spend more the 4% of GDP and more than 15% of the public sector budget on education, namely Costa Rica and South Africa, but the vast majority is not there, not even close. Redirecting scarce resources toward AI implementation before these foundations exist represents a category error—addressing a second-order problem while first-order failures persist.

The China Comparison and What It Obscures

Advocates point to China's AI deployment in schools as proof of concept. But this comparison obscures more than it reveals. China is an upper-middle-income country with near-universal electricity access, extensive internet infrastructure, and domestic technological capacity. Their AI initiatives build atop a functioning education system with adequate teacher supply, reasonable class sizes, and established curricula.

More importantly, we lack rigorous evidence that China's AI deployments actually improve learning outcomes for students performing below proficiency. Pilot programs and corporate announcements differ substantially from peer-reviewed impact evaluations comparing AI-assisted instruction to well-implemented traditional methods for struggling learners.

The countries that have begun introducing AI in primary education—China, UAE, Singapore—share a critical characteristic: they've already solved the problems that plague LIC/LMIC systems. They have electricity, internet, trained teachers, manageable class sizes, and adequate facilities. Their AI experiments explore enhancement, not remediation of system failure.

Low-Cost Training Programs: A Familiar Mirage

The proposal for "low-cost training programs" to teach teachers AI use deserves particular scrutiny. Training a new teacher requires 2-4 years and represents massive sustained costs. What evidence suggests that AI pedagogical training can be delivered cheaply and effectively when traditional pedagogical training struggles?

Teacher professional development in LIC/LMIC contexts faces well-documented challenges: cascade models that dilute quality, one-off workshops without follow-up support, training disconnected from classroom realities, and no time or incentive structures for implementation. Adding AI tools to this broken system won't fix it. Instead, it adds complexity that requires ongoing technical support, device maintenance, content updates, and troubleshooting—all requiring sustained investment.

The pattern is familiar from previous technology interventions. Laptops distributed without teacher training or technical support. Computer labs built but never used because teachers lack skills or confidence. Educational software in English for students who don't speak English. The graveyard of ed-tech pilots in developing contexts should instill humility about what "low-cost" technology interventions can achieve.

What Would Actually Work

If we're serious about improving learning outcomes for the majority of children in LIC/LMIC countries currently achieving below minimum proficiency, the evidence points elsewhere.

First, address the teacher knowledge gap directly. Many teachers in these contexts never mastered the content they're meant to teach. No technology substitutes for ensuring teachers themselves understand basic mathematics and literacy. This requires serious investment in pre-service and in-service training focused on subject knowledge, not just pedagogy or technology use.

Second, reduce class sizes to manageable levels. The research is unambiguous that individual attention matters for struggling learners. A teacher cannot effectively support 60+ students regardless of technological augmentation. This requires the boring work of hiring, training, and retaining millions more teachers—exactly what the $97 billion financing gap represents.

Third, ensure adequate instructional time and materials. Many schools operate partial schedules because teachers juggle multiple jobs to survive on inadequate salaries. Students lack books. These are solvable problems that don't require technological innovation—they require political will and sustained financing.

Fourth, invest heavily in early childhood development. By age 10, when most students are functionally illiterate and innumerate, the damage is largely done. The research consistently shows early intervention is far more cost-effective than later remediation. Yet early childhood receives a tiny fraction of education budgets.

Conclusion: Avoiding Comfortable Evasions

The question posed—under what conditions could AI-assisted instruction cost-effectively improve learning outcomes for struggling students in LIC/LMIC contexts—has a uncomfortable answer: under conditions that don't currently exist and won't exist by 2030.

AI might play a useful augmentation role once basic infrastructure exists, teachers possess adequate subject knowledge and pedagogical skills, class sizes are manageable, and students have foundational early learning. Using it as a substitute for these investments represents the same political evasion that shaped SDG4 targets in the first place—the fantasy that technology can bypass the hard, expensive, slow work of building functional education systems.

The learning crisis is real. The SDG4 targets were indeed chosen for political palatability rather than technical necessity. And AI will not solve the problems we're avoiding. The sooner we admit this, the sooner we might actually fund what works.

References

1. UNESCO. (2023). Global Education Monitoring Report 2023. Retrieved from gem-report-2023.unesco.org

2. United Nations. (n.d.). Sustainable Development Goal 4: Quality Education. Retrieved from sdgs.un.org/goals/goal4

3. World Economic Forum. (2023). These are the barriers facing global education targets: UNESCO. Retrieved from www.weforum.org

4. United Nations. (n.d.). Goal of the Month – Goal 4 – Quality Education. Retrieved from www.un.org/sustainabledevelopment

5. United Nations Statistics Division. (n.d.). SDG Indicators – Goal 4. Retrieved from unstats.un.org

6. United Nations Statistics Division. (n.d.). SDG Indicators – Goal 4 (School Infrastructure). Retrieved from unstats.un.org

7. World Economic Forum. (2023). These are the barriers facing global education targets: UNESCO. Retrieved from www.weforum.org

8. UNESCO. (2021). Global Education Monitoring Report 2021/2. Retrieved from gem-report-2021.unesco.org

9. World Bank. (n.d.). Country Classifications.

10. United Nations. (n.d.). Goal of the Month – Goal 4 – Quality Education (Teacher Training). Retrieved from www.un.org/sustainabledevelopment