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
2024-02-01
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.
Computers and Society
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: How to use large - language models (LLMs) to help teachers improve teaching quality in resource - poor schools. Specifically, the paper focuses on the education system in Sierra Leone, a low - income country, and explores how to provide support and training for teachers through an AI chatbot (TheTeacher.AI) to improve teaching effectiveness. ### Core issues of the paper 1. **Lack of educational resources**: - In developing countries, educational resources are limited, especially in poor areas. For example, many schools in Sierra Leone lack basic facilities such as electricity and the Internet, and teachers often have not received professional training. 2. **Insufficient teacher support**: - The main challenges faced by teachers include large - class teaching, lack of opportunities for professional development, and difficulty in obtaining support for classroom teaching. Especially in remote areas, teachers have even more limited opportunities to obtain support. 3. **Feasibility of technology application**: - Although large - language models (LLMs) have shown great educational potential in wealthy areas, there are still many obstacles to their application in low - income countries, such as unstable network connections and limited devices. ### Goals of the paper The paper aims to explore the following questions: - How to design and implement an AI chatbot suitable for the educational environment of low - income countries? - Can this chatbot effectively support teachers' professional development and classroom teaching? - Can the use of AI chatbots improve teachers' teaching quality and students' learning outcomes? ### Research methods To answer these questions, the research team developed TheTeacher.AI, an AI chatbot based on GPT - 3.5 Turbo, which interacts with teachers through WhatsApp. The research team evaluated in the following ways: - **Preliminary implementation**: Early deployment was carried out in 122 schools and among 193 teachers. - **Qualitative observation**: Analyze teachers' usage and feedback. - **Data analysis**: Analyze 6,900 query data to understand how teachers use the system and its effects. ### Conclusions Research shows that, despite technological and infrastructure limitations, AI chatbots can still help teachers improve teaching to a certain extent. Especially for those resource - poor schools, AI tools can provide immediate support and guidance, helping teachers better plan courses, manage classrooms, and answer subject - related questions. However, to make these tools truly effective, technological barriers need to be overcome, and teachers need to be provided with sufficient training and support. --- In summary, this paper attempts to solve how to provide effective support for teachers through AI technology in a resource - poor educational environment, thereby improving teaching quality.