Today, we’re sharing the compelling results of a recent randomized controlled trial (RCT) in Sierra Leone, exploring AI’s impact on student learning. This eight-week study, partnered with Fab AI and supported by the Sierra Leone Ministry of Education, reveals exciting possibilities for educational technology.
The trial focused on 1,763 junior secondary students across 12 schools in the Port Loko District, evaluating how “Guided Learning” in Gemini influenced their math progress. The findings strongly suggest that carefully designed AI can significantly enhance learning outcomes, powerfully augmenting dedicated teachers rather than replacing them.
A New Era for Learning: AI as a Pedagogical Partner
The Sierra Leone Ministry of Education, keen on innovation yet demanding rigorous study, expressed delight in this strong evidence. Their endorsement validates AI as a powerful tool to augment and extend the reach of hardworking educators, not replace them.
A common concern is that generative AI could become a shortcut, bypassing the challenging cognitive effort essential for deep learning. Our “Guided Learning” approach, built from years of LearnLM research, addresses this by prioritizing understanding over providing direct answers.
This pedagogically-grounded design engages students in a “Socratic” manner, prompting critical thinking. Analysis of over 113,000 interactions showed students built conceptual understanding in an impressive 91.4% of conversations. Gemini posed scaffolding questions in 76% of its messages, providing direct solutions in only 2% of cases, maintaining active student engagement.
Unpacking the Impact: Engagement and Deeper Understanding
This trial’s success stemmed from a strong partnership between advanced AI and dedicated educators, who remained firmly in control. Teachers meticulously designed lessons, set clear objectives, and facilitated dynamic classroom discussions, driving learning forward.
Teachers reported Gemini also supported their professional growth, using it for lesson preparation to discover new ways to explain topics like fractions. Many shifted from “lecturers” to “facilitators,” guiding students through personalized learning journeys.
Student engagement was remarkably high, defying typical educational technology adoption rates. A staggering 69% of students met or exceeded usage targets, far surpassing the typical “Five Percent Problem” for voluntary ed-tech engagement.
This high engagement translated into a profound shift in student behavior. Students reported enjoying math more and actively engaged beyond regular lessons. Crucially, their questions evolved, becoming more learning-oriented and focused on skill-building rather than seeking direct answers.
Skill-building queries rose to 90% by the final week (up from 68% in week one), while solution-seeking questions dropped from 25% to just 10%. This showed students desired understanding, not just answers.
Empowering Educators and Boosting Outcomes
The quantitative results were truly significant: students using Guided Learning saw a gain of +0.258 standard deviations in their math scores compared to the control group.
Practically, this represents roughly 1.2 to 1.7 years of typical learning progress achieved within just eight weeks. Classrooms where teachers integrated Gemini into approximately half their lessons (targeting 12 hours) saw even higher gains: roughly 1.8 to 2.5 years of academic progress.
To help others implement similar impactful programs, we are releasing a comprehensive teacher training guide, including materials created with Fab AI. It outlines the specific protocols used in this study, making replication and adaptation easier.
The Path Forward: Addressing Gaps and Scaling Success
While promising, these results also highlighted the “achievement gap”: stronger students benefited most, even as the majority gained significantly. This underscores the critical need to develop AI tools that deliver the strongest gains for students who need them most.
To advance open science and disseminate timely insights, we’re also releasing a playbook detailing our approach to RCTs with Fab AI. This guide aims to empower others to conduct faster, scalable studies, fostering robust, localized evidence that keeps pace with technological advancements.
We will expand these trials to other countries, delving deeper into metacognition and relational intelligence for a holistic understanding of learning’s complexities. By combining teacher-led classrooms with AI’s personalized scaffolding, our goal is to ensure technology serves as a vital bridge to meaningful learning for every student.
- We acknowledge additional support from Google.org and the Gates Foundation.
- Collaborators also included EducAid, Laterite, and Oxford MeasurEd.
- Our ongoing support for the Global AI for Learning Alliance (GAILA) will further accelerate these commitments through collective action.
Source: Google DeepMind Blog
