Opening Remarks — Parallel Session 1.4 (delivery: ~2.5 min)
Good morning, colleagues and dear participants in this room and online,
It is a pleasure to welcome you to our Session on "Frugal AI for equitable learning: Small models, offline solutions and wider access."
AI-supported education is often imagined in conditions of digital abundance — reliable connectivity, powerful computing, and ready access to devices, skills and support. But what changes when those conditions cannot be taken for granted?
This is the central question for today’s session: how can AI support learning where resources, connectivity and infrastructure are limited—and how can we ensure that it expands opportunity rather than deepens existing inequalities?
This matters for three reasons:
First, equity. If effective AI-supported learning depends on infrastructure that many learners and education systems do NOT have, AI risks reinforcing educational divides rather than closing them.
Second, design. Constraints are not only technical — they involve cost, access to teachers, digital skills, and the realities of different communities. Designing around those realities from the outset makes solutions more accessible, relevant and robust.
Third, innovation. Designing for real constraints leads to different architectures: smaller, more focused models, offline-first approaches, and new forms of community-driven learning.
This is closely connected to UNESCO’s work on green and resource-efficient AI. Our joint UNESCO–UCL research, Smarter, Smaller, Stronger, highlights how in some cases, smaller, task-specific models can use up to 90 per cent less energy than larger, general-purpose alternatives.
The implication is important: the future of AI can be shaped by resilient, right-sized and efficient AI systems—systems better adapted to low-resource and infrastructure-constrained settings, andbetter aligned with environmental sustainability.
UNESCO is working to translate this evidence into practice. Through our Green AI capacity building programme for policymakers and regulators; and through open innovation initiatives such as our Resilient AI Challenge with Governments of France, India, and tech partners such as Google, Mistral and Sarvam AI., we are working to turn that research into practice.
Today, we will see how these questions play out in education. We will move from an equity-centred perspective rooted in the Global South, to offline-first foundational learning, to concrete approaches in mathematics and STEM, and finally addressing bigger digital divide questions, through a community-driven digital learning presentation.
The contexts and technologies differ, but they share a common ambition: to design AI-supported learning around the real needs, resources and realities of learners and communities.
With that, let us dive into the first panel discussions.
AIED Unplugged: Practices, challenges and pathways forward in the Global South
Rafael Ferreira Mello: Rafael Ferreira Mello is a Professor and Chair of Education Technology Adoption and Data Analytics at CESAR, Brazil, and an affiliated researcher at the UNESCO UNITWIN Chair on AI in Education Unplugged. He leads educational technology R&D funded by the Brazilian Ministry of Education, Google, and Cogna Education, with partners across Europe, Australia, and Latin America.
Marcela Lorenzoni: Operations Coordinator, UNESCO Chair on AIED Unplugged, Brazil