Preparing classrooms for responsible use of AI: Insights from GSF AI Roundtables

Over the past few months, Global Schools Forum (GSF) convened education practitioners to examine the use of AI in education through three lenses: safeguarding, AI literacy, and instructional design and assessment. The discussions surfaced the reality that while AI is already part of practitioners’ daily work, schools and education systems may need more preparation to support its effective, safe and responsible use.

A key question emerged from these discussions:

What does responsible, locally useful adoption of AI look like when infrastructure, policy and teacher training are still catching up?

AI is already here and shaping students’ lives through the devices in their pockets, teachers’ WhatsApp groups, submitted essays, and lesson plans drafted late at night.

Teachers and organisations are already experimenting, although they are doing so around major structural gaps. During the conversations, practitioners shared various ways they are designing and encountering AI in their work across assessments, instructional design and AI literacy and safeguarding. In Rwanda, organisations have embedded AI into learning management systems for grading and tutoring. In Kenya, AI-generated lesson guides are being sent to teachers through WhatsApp to work around unreliable broadband. In Afghanistan, AI is being used to create Bloom’s Taxonomy-aligned assessment questions for girls excluded from formal secondary education. In India, AI is being tested for foundational literacy across regional Hindi dialects.

This is not a sector standing on the sidelines. It is a sector making do with the tools available, often before the right tools have been built for its context.

The problem is not resistance. It is infrastructure.

A simple story would blame slow adoption on reluctant teachers or cautious institutions. The roundtables do not support that view.

The barriers raised again and again were practical and structural: poor connectivity, limited access to devices, the cost of mobile data falling on teachers, and uncertainty around data ownership and accountability.

In India, one promising literacy assessment pilot did not stall because the technology failed. It stalled because there was no clear answer on who owned the data collected in government schools under the Digital Personal Data Protection Act.

That is not a training problem. It is a governance problem. Teacher enthusiasm cannot solve it on its own.

Human judgement is becoming the rule, not the exception.

One of the clearest points of agreement was that AI should not be left to make decisions unchecked.

Across the sessions, organisations described some version of human-in-the-loop practice as essential. Not as a safety disclaimer, but as part of the design. In the instructional design session, curriculum specialists described AI as a “substitute mind”: a thinking partner that can test whether a question fits a lesson plan. But the final judgement stayed with the teacher or specialist. In the assessment session, teachers in Afghanistan allowed AI to generate multiple-choice distractors, but set the correct answers themselves. They treated automated marking as a risk to manage, not a shortcut to celebrate.

That caution is not anti-technology. It is a practical response to the known weaknesses of AI: hallucination, weak pedagogical alignment and content that sounds fluent but lacks cultural relevance.

Equity is not a side issue. It shapes everything.

The roundtables were clearest when they acknowledged that AI does not enter unequal systems neutrally. It often widens the gaps already there.

Paid AI tools usually perform better than free ones, but many organisations in low- and middle-income countries cannot afford them. That pushes schools and NGOs towards tools that may be less accurate, less culturally appropriate or less safe. Better-resourced schools are more likely to get governed, supported AI tools. Rural schools and low-fee schools are more likely to get whatever is free and available.

That is why safeguarding and AI literacy cannot be treated as separate agendas. A student using a shared family device, with limited digital literacy, is not only more exposed to unsafe content. They are also less able to build the skills needed to use AI well.

Protection and capability are two sides of the same problem. Policy that tackles one and ignores the other will struggle with both.

The biggest gap is evidence.

The hardest question remains unanswered: is AI improving learning? Right now, teacher time saved is the easiest thing to measure, but that is not enough on its own. A tool can save time and still make little difference to what students understand, remember or can do.

The assessment roundtable made an important distinction here. The strongest AI use cases at this stage are not in high-stakes exams. They are earlier in the process: question design, rubric development and formative feedback. In those settings, an AI mistake leads to a redraft, not a wrongly scored child. That is a cautious conclusion, but probably the right one.

The real finding is the convergence.

Taken separately, the three roundtables could look like a set of useful but scattered observations. Taken together, they show something more important: practitioners across more than a dozen countries are reaching very similar conclusions, despite working in very different settings. They are not rejecting AI, and they are not rushing blindly into it either. Instead, they are asking for something more practical: shared frameworks, trusted tools, stronger governance, honest evaluation and space to learn from one another.

The gap between AI enthusiasm and real educational value will not close by itself. It needs structure, evidence and trust. That is where a community like GSF’s can play a serious role.

That is where a community like GSF’s can play a serious role.

GSF is now putting that role into practice through a set of initiatives designed to help members move from early experimentation to more confident, responsible adoption.

We’re developing an AI and Education microsite to curate useful publicly available resources, including AI events, tools, case studies, guides, national AI in Education policies, AI literacy materials and safeguarding frameworks.
We’re launching an AI Leaders Circle, a peer learning group for school-system leaders, developed in collaboration with Transcend.
We are strengthening our AI Safeguarding Policy work with the GSF community, supporting organisations to update their Child Safeguarding toolkits to cover AI use cases.
We will run a series of Practitioner Spotlights to surface practical examples from across the network, including community-led and community-produced AI experiments.

GSF’s goal in this space is not to present AI as a finished answer. It is to create the shared space, trusted resources and practical evidence that help education organisations make better decisions together.

Share this article