Dr Paul Atherton is the Executive Director of Fab AI, a non-profit which brings together the work of Fab Inc. – an education and international development advisory firm, and AI-for-Education.org – a platform to guide the development of AI for education for low- and middle-income countries (LMICs)
Last year, in 2025, Fab-AI joined the EiE Hub as a member, with the aim of supporting the wider network of education actors in crisis situation to explore how AI can be used to both deliver education solutions, and to support teachers and students learning in emergencies.
In this Q&A, Paul discusses the work of his own organisations, how AI can help both children in emergencies and humanitarian actors, and why he decided to base himself with the EiE Hub at the heart of International Geneva.
Could you tell us about the work that your non-profit, AI-for-Education.org, does?
There are essentially three areas we work in. The first is related to quality assurance, where we test and benchmark different AI models to see how useful they are for education. This involves, for example, testing the pedagogical potential of different large language models (LLMs) to see if they can pass teacher exams, including how they fare on questions for SEND (special educational needs and disabilities) pedagogy
We’re also doing more specific evaluations to see how well LLMs are adapted to be learning tools. For example, can an LLM generate a story including CVC (consonant-vowel-consonant) words that are critical for early readers, or that introduces new concepts in a way that doesn’t overwhelm children? Are AI tools adapted to local contexts? If an LLM generates a lesson plan for Tanzania centered on pizza instead of chapati, for example, it’s already failed to meet learners where they are. We have partnerships with Google and the Gates Foundation to do this, and the results are available to anyone on our website.
We’re also trying to support at-scale implementation of AI. That is, how do we go from a smaller scale to the national level with some of these tools – and should we? We’re about to start doing pilot projects in Sierra Leone and Kenya to test how to do this responsibly. Finally, we’re also looking at climate smart school planning. This includes working on early warning systems, which in turn will help schools adapt to the climate crisis and improve its infrastructure.
Why should LMICs invest in AI for education?
Because you’ve got the entire world’s knowledge base accessible to anybody! Anyone, anywhere in the world can essentially get PhD-level answers at the tip of their fingertips. For teachers, it can help develop lesson plans, tests or a range of other solutions. For children, the benefits could be immense, as long the access, motivation to learn and ability to read is there. AI will never replace in-classroom learning, but as a tool to support education it can be invaluable – even if it just involves spending an extra 30 minutes with an LLM at the end of the day to supplement learning.
The problem we’re facing now is that most AI-enabled EdTech products are tailored to higher-income contexts, where the learning conditions and availability of infrastructure are much different to LMICs, where the greatest needs are. This is changing, however, as more and more EdTech products using AI are being implemented in LMICs, and early evidence is pointing to some real success stories.
What about in humanitarian or crisis settings – how can AI help education delivery there?
Several ways. One of the key challenges in crisis settings is disrupted schooling – children who are forced to flee their home won’t be able to go to school every day, for example. There is potential for AI to help plug this gap, since it works anywhere and doesn’t need access to a physical school. There are several organisations, including IRC, already doing interesting work on this. It can also support home learning. In Afghanistan, for example, we helped connect a local partner that is supporting girls learning at home – since they have been banned from attending schools – to some of our engineers to help them build a solution. With the funding crisis affecting the humanitarian sector, there are of course also costing benefits, as AI can help improve efficiencies in operations and develop learning tool.
AI is developing and being adopted at a breakneck pace, and there are understandably concerns about potential pitfalls. Do you see any particular risks in using AI for education, and how do you mitigate those?
Of course. There are risks around the content produced by LLMs, which often relate to the quality of the training data. This can lead to biases in answers or answers that are simply wrong. That is why we put such an emphasis on testing different models and ensuring quality control. There are also potential cognitive risks for children who use AI in learning. The key here is to test extensively in a safe, controlled environment, and also to involve and talk to teachers about what they are observing. Finally, there are also safety risks in terms of what content children might be able to access. This is where guardrails, standards and legislation are so important.
Last year, Fab AI moved into the EiE Hub offices in Geneva. Why did you decide to set up shop here?
The office is just a really nice and welcoming space, full of very committed people. It’s nice to have access to that network of very high-quality people. I find it genuinely useful to come into the office every week! Geneva is also an international hub, of course, with access to many organisations that are key to setting international standards, and are central to a whole range of sectors. And Switzerland is actually also a quiet power when it comes to AI – there’s a lot of great work coming from here, such as from ETH Zurich, Swiss Federal Technology Institute of Lausanne, and from the big labs.