DeepMind AI Trial Boosts Math Scores in Sierra Leone by 1.7 Years
Google DeepMind has released results from a groundbreaking randomized controlled trial in Sierra Leone, showing how AI-powered tools can significantly improve math education. Students using DeepMind’s Guided Learning system achieved gains equivalent to 1.2 to 1.7 years of typical learning in just eight weeks. Classrooms where teachers used the tool extensively saw even higher progress, with gains reaching 1.8 to 2.5 years.
The trial, backed by Google.org and the Gates Foundation, centered on Guided Learning, a system designed to augment, not replace, teachers. Guided Learning prioritizes conceptual understanding and critical thinking over rote solutions. According to DeepMind, 91.4% of student interactions with the AI focused on building understanding, while direct answers were provided in only 2% of cases. The system’s "Socratic" approach encourages students to solve problems themselves, supported by scaffolding questions in 76% of AI responses.
AI in Action: Teacher-Led, Student-Centered
Rather than replacing educators, the trial emphasized the importance of teacher involvement. Lesson design and classroom facilitation remained teacher-led, with AI serving as a support tool. Teachers reported not only improved student engagement but also professional growth, discovering new methods to explain challenging topics like fractions.
The trial also addressed a persistent issue in educational technology: low engagement. While similar tools often see participation rates below 5%, 69% of Sierra Leonean students met or exceeded usage targets. This suggests that Guided Learning not only improved learning outcomes but also made math more enjoyable and engaging for students.
Scaling Global Impact
This trial is part of a broader push to explore AI’s potential in education globally. The initiative aligns with recent moves by organizations like UNESCO and UNICEF to incorporate AI into school systems in developing nations. For example, the Philippines recently committed to training over one million educators in AI skills, while UNESCO launched an AI education observatory in Latin America and the Caribbean in April 2026. These efforts reflect a growing consensus: AI has the potential to bridge gaps in teacher shortages and resource constraints if deployed equitably and responsibly.
However, the Sierra Leone trial also highlighted a key challenge: the "achievement gap." Students with stronger math skills benefited the most, underscoring the need for tools that can better support lower-performing students. DeepMind plans to address this by refining its tools and expanding trials to additional countries, focusing on areas like metacognition and relational intelligence to capture more nuanced learning improvements.
Implications for Policy and Practice
Education policymakers in developing countries are increasingly looking to AI as a solution to systemic challenges. The World Bank’s 2026 report on AI for development emphasizes the technology’s potential to transform education but warns of barriers like infrastructure gaps and limited teacher training. The success of initiatives like DeepMind’s trial will depend on how well these challenges are addressed.
To support broader adoption, DeepMind has released training guides and a playbook for conducting randomized controlled trials in diverse contexts. These resources aim to accelerate evidence-based deployment of AI tools in education while fostering global collaboration through partnerships like the Global AI for Learning Alliance (GAILA).
Looking Ahead
AI’s role in education is rapidly evolving, with DeepMind’s Sierra Leone trial offering a compelling case for its potential. But as the trial showed, success requires more than just technology—it depends on teacher leadership, equitable implementation, and an unwavering focus on student learning. With plans to expand trials globally, DeepMind’s efforts may provide a roadmap for how AI can transform education systems, particularly in resource-constrained settings.