How are education implementers approaching AI content generation, curation, and integration?

This blog discusses the conversation from a Community of Interest (COI) event on Artificial Intelligence in Education, a collaboration by the World Bank and EdTech Hub.

The emergence of Generative AI is challenging traditional approaches to the development of teaching and learning materials in education systems around the world. AI-powered tools are transforming traditional models of curriculum design and assessment, enabling rapid generation of adaptive, localised learning content, and automating resource creation to address gaps in access, language, and quality. 

Many of the most innovative approaches to utilising AI in education are emerging across Asia, part of larger efforts to reimagine the development and use of digital teaching and learning resources at scale across education systems. The World Bank and EdTech Hub’s AI Observatory brought together three different experts to discuss their different approaches and insights from their work on this topic — and more. 

We’re sharing what we heard during this special session on 16 July 2025, and what questions remain from our special COI convening on the changing role of Generative AI in education. 

Featured Speakers 🔉

• Verna Lalbeharie, Executive Director, EdTech Hub
• Mike Trucano, Senior Advisor, EdTech Hub AI Observatory and Action Lab and Non-Resident Fellow, Brookings Institution
• Mark West, Education Specialist, UNESCO
• Joleen Liang, Co-Founder, Squirrel Ai Learning 
• Srikanth Talapadi, Managing Director,  Chimple Learning 

Watch the webinar 📺

Catch up on the full conversation.

Key takeaways 📝

Here are four key takeaways from the discussion about what works and what does not from AI-generated content for education, particularly in the context of low- and middle-income countries:

1. There are diverse approaches to AI-supported content development and application

The session revealed different approaches to how AI can support educational content development. Squirrel AI focuses on learners engaging independently with content originally developed by curriculum experts but dynamically adapted by AI to fit their personal learning needs and behavioral patterns. Their Large Adaptive Model (LAM) personalises existing high-quality content created by experienced teachers, allowing students to follow unique learning trajectories through pre-designed materials. In contrast, Chimple places current teachers at the center of content generation, providing them with AI workbenches that help generate engaging, gamified content tailored to their specific classrooms, communities, and cultural contexts. These contrasting approaches —AI as content personaliser versus AI as teacher empowerment tool—represent key design choices for how artificial intelligence can support educational content development.

2. Consider adopting a “Context In, Context Out” approach to tackling the challenge of content localisation and relevance

As Srikanth from Chimple articulated, AI-generated educational content quality depends heavily on contextual relevance across five levels: language, region, community, classroom, and individual child. Most large language models are trained primarily on English and Western content, and can include biases introduced through their training. To help relevance and utility, children in rural India need content that reflects their lived experiences—using familiar foods like rice instead of pizza in math problems, incorporating local stories, and addressing specific classroom dynamics. One way to address this challenge is to start the content development process with input from local teachers as Chimple has done to ensure that the end-product developed by AI is contextually relevant.

3. Quality control and trust are fundamental to educational integrity

All speakers emphasised the critical importance of content validation and quality assurance. Mark West from UNESCO warned against AI’s tendency to produce “bland, generic, average, anodyne” content that can represent cultural hegemony and take the place of content that is more on the fringes, potentially offering more interesting and engaging perspectives. Both Squirrel AI and Chimple have developed different approaches to quality control—through evaluation agents and teacher verification layers respectively. It was broadly seen that AI-generated “slop” cannot be allowed to infiltrate education systems, as this would erode trust between schools, parents, and communities.

4. Defining personalisation of content and engagement is not necessarily straightforward

The session revealed that “personalisation” in AI-generated content is far more complex than individual student preferences. Squirrel AI demonstrated personalisation through behavioral data and adaptive learning paths, creating over 1 million unique learning trajectories. Chimple approached personalisation through a five-level framework: language, region, community, classroom, and individual child. Mark West challenged the field to think beyond individual personalisation, noting that governments may want personalisation at classroom, school, or national levels to maintain shared educational experiences rather than individual. The discussion highlighted tension between AI’s personalised paths (which often put a high value on efficiency) and learners’ need for exploration and exposure to diverse perspectives.

Questions 🙋🏾‍♀️

The following questions were posed from community members. We’re sharing to help stimulate further discussions and knowledge exchanges. Please note some questions may have been edited for spelling or clarity.

Resources 📚 

🔎 The following resources were shared by community members and participants. These have not been reviewed by the World Bank or EdTech Hub, but are useful indicators of what conversations, evidence, and methods are being explored in the sector. 

---

This is part of an on-going series hosted by the World Bank and EdTech Hub’s AI Observatory and Action Lab. The AI Observatory is made possible by support from UK International Development. Please follow along and join the conversation on LinkedIn!