Are LLMs Useful in the Poorest Schools? TheTeacher.AI in Sierra Leone
Jun Ho Choi, Oliver Garrod, Paul Atherton, Andrew Joyce-Gibbons, Miriam Mason-Sesay, Daniel Björkegren
TL;DR
The paper investigates whether large language model technology can support teachers in the world’s poorest classrooms by introducing TheTeacher.AI, a GPT-3.5 Turbo–based WhatsApp chatbot tailored to Sierra Leone. It reports a field pilot across 122 schools with 193 teachers, using qualitative observations and query analysis to assess usage, patterns, and local adaptation. Findings indicate sustained engagement over a school year, with a subset of teachers using the tool more regularly and using it mainly for concept clarification and lesson planning; system tailoring to local conditions significantly improved perceived usefulness. Key challenges include unreliable connectivity, limited device access, high inference costs, and safeguarding concerns, underscoring the need for offline modes, human oversight, and further evaluation before scale-up. The work suggests AI-assisted tools can benefit low-resource schools when designed for local constraints and accompanied by training and oversight, though implementations will differ from those in wealthier settings.
Abstract
Education systems in developing countries have few resources to serve large, poor populations. How might generative AI integrate into classrooms? This paper introduces an AI chatbot designed to assist teachers in Sierra Leone with professional development to improve their instruction. We describe initial findings from early implementation across 122 schools and 193 teachers, and analyze its use with qualitative observations and by analyzing queries. Teachers use the system for lesson planning, classroom management, and subject matter. Usage is sustained over the school year, and a subset of teachers use the system more regularly. We draw conclusions from these findings about how generative AI systems can be integrated into school systems in low income countries.