In just six weeks, secondary school students in Edo, Nigeria participating in an after-school program combining AI tutoring with teacher guidance, achieved learning gains of 0.31 standard deviations, equivalent to roughly 1.5–2 years of typical schooling. Behind these results was a thoughtfully designed intervention that placed teachers and students at the center, harnessing the promise of generative AI while drawing from critical lessons from past EdTech initiatives, particularly during the COVID-19 pandemic.

As an education specialist and co-lead of the World Bank’s EdTech Team, I have spent the past few years pioneering responsible AI-powered education programs in Nigeria, Peru, Brazil, and beyond. My experience has shown that the key to success is not the technology itself, but a holistic approach to program design: one that starts with the educational problem, considers the local context, and always puts people first.

In this blog, I share five practical insights to support policymakers and education leaders towards more responsible and effective integration pathways of generative AI in education. These lessons offer a framework to shape the so-called "AI revolution" to better address the global learning crisis.

1. Start with the educational problem, not the AI tool

Before considering AI tools as a potential solution, we must clearly define the educational challenge we aim to address. Are we seeking to improve teaching quality, provide targeted support to students falling behind in math, or enhance digital literacy? Simply incorporating AI does not guarantee usage or impact.

This aligns with the World Bank's "Ask Why" principle: EdTech policies and projects must be grounded in a clear vision for educational change. Technology should be a means to an end, not the end itself. This ensures we do not adopt technology for its own sake, but as a targeted tool to solve a specific problem.

2. A focus on the purpose of the program: develop a strong theory of change

Effective program design should start with a clear theory of change, which is a description of how an intervention is supposed to deliver the desired changes, outlining the causal pathways and assumptions involved. In other words, it clarifies the link between activities, outputs, outcomes, and the ultimate goals. For example, teacher training (activity) may lead to new instructional practices (output), which in turn improve student engagement (outcome) and lead to learning gains (impact).

Having a strong theory of change will not only support a clear vision of how the AI tool may support the intended purpose of the program but also highlight the key activities and outputs needed for that to happen (for instance, teacher training, guidelines on AI literacy, focus groups with students or parents, etc.) and establish indicators that measure use, progress, and engagement.

Moreover, a well-articulated theory of change provides the foundation for impact evaluation, helping to build the evidence base on what works and what does not when integrating AI in education.

3. Context is critical: design for the needs of the place

Successful integration of AI depends on understanding the local context. Factors like connectivity, electricity, and device availability should shape the choice of tools and content. For example, in areas with limited devices such as laptops or desktops or intermittent Wi-Fi, using mobile phones can be a more suitable option, as seen in an initiative in Ghana.

Context should also inform content and tool selection. For example, training materials for teachers and students in Nigeria and Peru were customized to their respective cultures and local realities, avoiding a one-size-fits-all approach and ensuring relevance and accessibility.

4. Technology in service of people

Education is fundamentally about human connections. The use of generative AI must be in the service of teachers and students, supporting their needs and focused on improving learning and teaching practices. This means involving stakeholders ranging from teachers and students to parents and school leaders to co-design programs. It also requires comprehensive AI literacy training, prioritizing learners’ needs, addressing concerns and limitations and fostering effective, inclusive communities of practice, among other considerations.

Last year in Lima, Peru, we held focus groups in public schools to understand teachers' current use of AI and how they envisioned it helping them in their daily work. This collaborative process led us to select innovative teachers who co-designed training materials with us and identified use cases where AI could enhance their teaching practices.

This people-centered approach ensures that technology empowers, rather than displaces, the human element in education.

5. Ensure the product is fit for purpose and context

Finally, any AI tool or solution must be suitable for the specific context and purpose. This involves a careful assessment of its features, including but not limited to:

• Safety and privacy: does the product have robust policies to protect user data?
• Cultural and linguistic appropriateness: is the tool available in local languages, sensitive to cultural norms and designed to avoid bias and stereotypes?
• Adaptability: can the solution be tailored to the specific needs of the educational environment and diverse learners?
• Optimized for learning: is the tool aligned with the science of learning principles, and has it been evaluated with intended users?
• Accessibility: are there offline options and support for learners with disabilities?

This "5 Ps" framework – Problem, Purpose, Place, People, and Product – provides a structured way to approach the design and implementation of AI in education programs.

A practical framework for responsible AI integration

To help guide program design, consider these key questions at each stage:

• Problem: what is the educational problem being targeted? would educational technology help? in what way? If so, would generative AI play a role?
• Purpose: what is the main goal of the program? Is there emerging evidence of previous use cases? would generative AI be in the background or user-facing? can you illustrate how AI is articulated in your theory of change?
• Place: what are the connectivity and electricity conditions? what type of devices are available? can AI tools be adapted to the context?
• People: who should be involved in co-design? what are their views on AI in education? Have they used AI tools or received AI literacy training?
• Product: is the solution fit for purpose and context? Are safety and privacy policies in place? is it suited to the local language and culture?

Conclusion

The debate around AI in education is often polarized, but responsible integration requires nuance and leadership. Educational leaders must commit not only to responsible use but also to the responsible design and implementation of AI programs. The AI revolution in education is not something to passively accept; it is an opportunity to actively shape solutions that address the global learning crisis.

By focusing on clear problems, articulating purpose, and goals, designing for the local context, and placing people at the center, we can build a framework for responsible AI integration that truly serves students and teachers. I invite you to share your experiences and join the conversation as we work together to harness AI’s potential for education.

This blog post was originally published on the Blavatnik School of Government at Oxford University’s “Voices” page.