Google DeepMind Guided Learning AI Trial in Sierra Leone

Google DeepMind has conducted a pre-registered trial in Sierra Leone demonstrating that AI can serve as a powerful pedagogical partner to augment teacher reach without replacing educators. The study found that students using Guided Learning achieved significant gains in math scores, with some experiencing the equivalent of 1.2 to 2.5 years of typical learning progress within an eight-week period.

Quantifiable Learning Gains and Student Engagement

Students using Guided Learning saw a gain of +0.258 standard deviations in math scores compared to a control group, representing approximately 1.2 to 1.7 years of typical learning progress. The impact was most pronounced in classrooms where teachers integrated Gemini into roughly half of their lessons to meet a target of 12 hours of usage during the trial; these students saw gains of roughly 1.8 to 2.5 years of progress.

Engagement levels significantly outperformed typical educational technology trends. While voluntary educational technology often suffers from "The Five Percent Problem" (where only 5% of students typically engage), 69% of students in this trial met or exceeded usage targets. This high engagement was accompanied by a student-reported increase in enjoyment of mathematics.

Pedagogical Approach: Prioritizing Understanding Over Answers

Guided Learning, built on LearnLM research, is specifically tuned to prioritize conceptual understanding over the provision of direct answers. This "Socratic" interaction model ensures that students perform the necessary cognitive effort for learning.

Analysis of over 113,000 interactions during the trial revealed the following:

• Conceptual Understanding: Students used the tool to build understanding in 91.4% of conversations.
• Scaffolding: Gemini responded with scaffolding questions in 76% of its messages.
• Direct Solutions: Gemini provided direct solutions in only 2% of cases.

Over the course of the trial, student behavior shifted toward skill-building. Skill-building queries increased from 68% in the first week to 90% by the final week, while solution-seeking questions decreased from 25% to 10%.

The Role of the Teacher in AI Integration

The trial was designed as a teacher-led intervention where educators remained central to the experience. Teachers designed the lessons, set the objectives, and facilitated classroom discussions.

Beyond student impact, teachers reported professional growth, using Gemini for lesson preparation and discovering new ways to explain complex topics like fractions. This shifted the teacher's role from a "lecturer" to a "facilitator," allowing them to provide targeted support to student pairs as they navigated their learning journeys.

To support the scaling of these programs, Google DeepMind has released a teacher training guide created in collaboration with Fab AI.

Limitations and Future Research

Despite the overall success, the trial highlighted a persistent "achievement gap," as students who entered the trial with stronger initial math skills benefited the most. This indicates a need for tools that provide stronger gains for the students who need the most support.

Google DeepMind is expanding these trials to other countries through additional pre-registered Randomized Controlled Trials (RCTs) and is collaborating with the Global AI for Learning Alliance (GAILA). To facilitate open science, they have released a playbook on their approach to RCTs with Fab AI to help other organizations run scalable, localized studies. Future research will explore metacognition and relational intelligence to capture a more holistic view of the learning process.