Enabling evidence-based education decision-making with agentic AI

The world faces a foundational learning crisis. Today, 251 million children around the world are out of school and nearly 70% of 10-year-olds in low and middle-income countries are unable to read or understand basic text. Tackling this gap in foundational learning is essential for achieving the United Nation’s Sustainable Development Goal 4 on ensuring inclusive and equitable education and promoting lifelong learning opportunities for all.

To address this challenge, Capgemini partnered with UNESCO and our ongoing collaboration with AWS to develop a solution based on Generative AI capable of analyzing large-scale learning assessment data, which provides rich insights but is often underused, in particular in low- and middle-income countries due to their constraints in time, expertise, finance, and infrastructure. The collaboration between UNESCO and Capgemini has the potential to enable countries to interpret the wealth of learning assessment data, facilitating evidence-based decisions that enhance learning outcomes, in this case, children’s reading capabilities.   

Driving innovation with competition

This project was a perfect fit for Capgemini’s Global Data Science Challenge (GDSC). By crowdsourcing the development process, the participants brought a tremendous variety of backgrounds and expertise while driving innovation with the power of competition. This approach had previously proven its value with successful efforts to apply AI for protecting biodiversity and studying oceanic environments.

“Every year, we’re excited to bring our collective experience to bear on an essential challenge,” says Niraj Parihar, I&D CEO, Capgemini. “Focusing in on education, we saw the potential Gen AI had to process data and help leaders around the world make more informed decisions when it comes to education policy.”

In the seventh Global Data Science Challenge (GDSC), participants were tasked with building an AI education policy expert capable of analyzing and providing insights from the PIRLS 2021 dataset, one of the world’s most comprehensive studies of children’s reading literacy. Almost 1,500 participants joined the challenge and developed AI systems that could intelligently answer complex questions about educational trends, student performance, and teaching practices.

By leveraging advanced large language models and agentic systems, the teams created solutions that could process and interpret data from 400,000 students across 57 countries to make sophisticated educational insights accessible to policymakers, educators, and researchers. The winning solutions demonstrated the ability to provide data-driven answers about everything from COVID-19’s impact on reading habits to 4^th graders’ reading habits and achievement, ultimately supporting UNESCO’s mission to ensure inclusive and equitable quality education for all.

Across Capgemini, teams used AWS technology, including Amazon Bedrock and SageMaker, to explore the possibilities. While learning to use the latest technology, teams collaborated and competed with each other to develop, train, and test their models.

 “UNESCO believes that AI, and, in particular, generative AI – if developed and deployed with a strong ethical foundation and a human-centered approach – can be a powerful tool to support policy makers, school leaders and educators in providing targeted assistance to learners,” says Borhene Chakroun, Director for Policy and Lifelong Learning Systems at UNESCO.

Breakthroughs in AI for education policy and practice

By the end of the GDSC, a panel comprised of representatives from UNESCO, AWS, and Capgemini selected a winning team after a thorough comparison of numerous models against the stated objectives. The chosen agentic AI solution demonstrates how to query and understand large datasets, combine this information with other verified sources of data, and explain the result both visually and as text.

“The teaching and learning experience are absolutely vital for the future of our planet and for global peace,” explains Mike Miller, Senior Principal Product Lead, Generative AI, AWS. “Throughout the competition, we’ve seen how generative AI can help make education more effective around the world and enable our global community to commit to more sustainable development.”

The winning team, insAIghtED, developed an innovative solution aimed at enhancing educational insights using advanced AI agents. Their model leverages the PIRLS 2021 database, which provides structured, sample-based data on reading abilities among 4th graders around the world.

However, recognizing the limitations of relying solely on this dataset, the team expanded their model to incorporate additional data sources such as GDP, life expectancy, population statistics, and even YouTube content. This multi-agent AI system is designed to provide nuanced insights for educators and policymakers, offering short answers, data visualizations, elaborated explanations, and even a fun section to engage users. 

The architecture of the solution involves a lead data analyst, data engineer, chart preparer, and data scientist, each contributing to different aspects of the model’s functionality. The system is capable of querying databases, aggregating data, performing internet searches, and preparing elaborated answers. By integrating various data sources and employing state-of-the-art AI frameworks like Langchain and crewAI, the insAIghtED model delivers impactful, real-world insights that go beyond the numbers, helping to address complex educational challenges and trends. 

“We’re unbelievably proud to be a part of this project. The GDSC is always an exciting opportunity, and this time was no different,” says Niraj Parihar. “The collaboration between ourselves, UNESCO, and AWS has showed the positive potential of Gen AI and taken a step in the advancement of a more data-driven decision-making in education.”

Already, this Generative AI solution can make data-based insights accessible to everyone, offering guidance previously available only to experts. This democratization of data and what it reveals has the capacity to guide action all over the world as policymakers can instantly access insights and evidence that help them make informed decisions about education.

Furthermore, UNESCO and Capgemini are committed to improving learning outcomes worldwide by releasing the solution under an open-source license. This will allow for global adaptation and use, contributing to a shared public good.