Improving foundational math learning in Jordan through AI-powered chatbots

Despite near-universal enrollment in primary education in Jordan, with over 1.8 million children attending basic education, learning outcomes remain low. In the 2022 edition of the OECD's Programme for International Student Assessment (PISA), Jordan ranked 73rd out of 81 participating countries in mathematics. This mirrors a broader global learning crisis: an estimated 125 million children worldwide lack basic literacy and numeracy skills despite years of schooling (World Bank, 2018).

The COVID-19 pandemic has exacerbated these challenges, particularly for students from disadvantaged backgrounds, including the approximately 7% of students in Jordan’s public schools who are Syrian refugees. Addressing these deep-rooted learning gaps requires innovative, scalable, and cost-effective solutions that go beyond traditional classroom methods.

Adaptive, personalized tools powered by artificial intelligence are showing promise in tackling these issues. They can tailor content to each student's level, ensuring they receive the right support at the right time, an approach known to be effective in improving learning outcomes (Banerjee et al., 2017; USAID, 2023).

Darsel’s innovation is an AI-powered mathematics chatbot that helps students practice math via commonly used messaging apps. The platform uses an adaptive algorithm to tailor questions to each student’s level and adjust content as they progress. Designed for low-bandwidth environments, Darsel works without requiring new devices or high-speed internet. Its 500,000+ curriculum-aligned questions, reviewed by education specialists, ensure pedagogical quality. The chatbot uses gamified features to make learning more interactive and boost student motivation and confidence. It also supports teachers by providing real-time usage reports sent via messaging apps helping them track progress and encourage usage, with optional in-class recommendations to reinforce learning.

The innovation has already been piloted at scale. In 2023, Darsel was adopted by over 2000 public schools in Jordan for Grade 7 mathematics, with more than 70,000 students using the platform. The pilote conducted on 7th grade found that students who used Darsel improved their math performance comparable to 3 months of additional high-quality instruction.

Building on this success, this project will evaluate the platform’s effectiveness for Grade 6 students across 6 districts in Jordan, involving over 20,000 students through a randomized controlled trial.

FID’s support will fund:

• The design and implementation of a large-scale randomized evaluation
• The development and testing of teacher support features
• Impact analysis on math skills, motivation, confidence, and career aspirations
• Dissemination of findings to policymakers and education stakeholders The project is led by J-PAL Middle East and North Africa, in partnership with Darsel and the Jordanian Ministry of Education, who has formally endorsed the initiative through a multi-year partnership agreement.

This project aims to provide rigorous evidence on the effectiveness of AI-powered chatbots for foundational learning and student well-being of Grade 6. Specific outcomes include:

• Improved math performance among grade 6 students in public schools
• Increased student motivation and confidence, particularly in mathematics
• Higher engagement among disadvantaged students, including refugees
• Evidence on the role of teacher engagement and personalized support in enhancing the effectiveness of EdTech interventions
• Insights on cost-effectiveness compared to other EdTech solutions The research will also examine whether the intervention can influence longer-term outcomes such as reduced grade repetition and increased interest in science, technology, engineering and mathematics careers.

This project will contribute to the global body of evidence on scalable EdTech solutions and provide valuable lessons for governments, donors, and implementers seeking to improve learning outcomes in low-resource settings.