Adaptive Teaching at Scale: AI Innovations to Address Learning Poverty

• In Event: AI-Enhanced Adaptive Teaching: From Evidence to Implementation

Proposal

Learning poverty in low- and middle-income countries is driven by structural challenges that education systems have struggled to overcome: instruction that does not sufficiently adapt to the needs of individual learners, teachers who often lack sustained support in large and diverse classrooms, and feedback systems that provide information too late to guide effective remediation. These bottlenecks fuel disengagement and inequality, reinforcing cycles of exclusion. Artificial intelligence, if carefully designed and responsibly deployed, offers promising opportunities to address these gaps by strengthening adaptive teaching and reinforcing teacher practice.
Evidence indicates that personalization can accelerate progress, especially for learners who start behind. A large-scale randomized trial in India found that adaptive platforms improved mathematics outcomes by 0.37 SD and Hindi outcomes by 0.23 SD after only 4.5 months, with the strongest effects for struggling learners. In Nigeria, experiments with generative AI tutors aligned to national curricula reported substantial learning gains, demonstrating the feasibility of differentiation at scale when paired with strong teacher engagement.
Teacher-facing AI tools also hold potential. In Sierra Leone, an AI-based chatbot provided practical support for lesson preparation and classroom management across more than 100 schools. In Ghana, locally fine-tuned AI models have been piloted to generate lecture notes, exercises, and assessments in national languages, offering contextually relevant materials that reduce teacher workload and complement existing pedagogy. Such tools can extend scarce coaching resources and empower teachers with actionable, curriculum-aligned guidance.
Feedback systems are another critical area. Traditional assessments often take months to reach classrooms, reducing their instructional value. Voice AI pilots in India and Kenya suggest that oral reading data can be captured and analyzed quickly, providing teachers with near real-time insight into learner progress. In Ghana, a low-bandwidth AI tutor operating via WhatsApp demonstrated the feasibility of continuous adaptive practice in mathematics, improving student learning growth in a matter of months.
The strategic challenge lies in scaling these innovations responsibly. AI cannot substitute for the fundamentals of structured pedagogy, effective teaching, and supportive school environments. Instead, its role is to reduce costs, shorten the cycle from policy to classroom practice, and make differentiated instruction more feasible at scale. To succeed, AI solutions must be co-designed with national partners, support local languages, function in low-connectivity environments, and uphold strong data privacy and child protection standards.
My contribution to this panel will examine how AI can serve as a practical enabler of adaptive teaching by addressing three systemic bottlenecks: personalization, teacher support, and timely feedback. Drawing on research and pilots from Ghana, India, Kenya, Nigeria, and Sierra Leone, I will argue that AI's greatest promise lies not in replacement, but in amplification, helping teachers and systems extend effective practices to all learners, especially the most marginalized. With careful safeguards and public-sector integration, AI can complement existing reforms and contribute to more equitable pathways out of learning poverty.

Author

• Jonathan Stern, Gates Foundation