From Pilots to Practice: Testing AI Governance Where Learning Happens

Authorship:

By Dr. Rhianna C. Rogers, Dr. Nan Travers and Mark Hooley, Global Inclusive Education Ecosystem (GIEE)

At a GIEE workshop with the Geneva Forum on 13 October, one participant offered a simple test for AI in schools: the machine measures, the human decides. It is easy to agree with, and much harder to build.

Education has seen many promising pilots that never scaled. A tool works in one classroom, the funding ends, and the lessons disappear. AI raises the stakes. Rules such as the EU AI Act now require human oversight of high-risk systems, but a rule on paper doesn’t show a teacher, school or regulator what good oversight looks like.

The Global Inclusive Education Ecosystem (GIEE) was created to close that gap. GIEE is a regulatory sandbox for AI in education: a protected space where educators, researchers, founders and policymakers test AI tools and governance standards together, before they are widely adopted.

Our approach has three steps. First, every case study is grounded in two shared frameworks. The GOER Trust Layer covers governance, outcomes, engagement and resilience, including data protection and checks for bias. The Global Learner Qualifications Framework (GLQF) tracks learners’ progress over time instead of relying on one-off, pass-or-fail tests. Second, each case study becomes a test of those standards in its own setting. Third, we compare the evidence across settings to see where the standards hold and where they need to change.

That comparison is the key. Our first cohort spans youth-led critical AI education, special educational needs, gender equity, knowledge mapping for fair assessment and career discovery. When the same auditable criteria are applied across such different settings, we are testing more than the tools. We are testing whether the governance standard itself works across countries, languages and institutions. Where it doesn’t, we adjust it with the people closest to the work.

Human in control, not just in the loop. Every AI-supported step has a defined point where a person can review and override the output, and participants keep control of their own data, including the right to delete it.

Measure less, but measure what matters. We cap the data we collect at what meaningful oversight needs, and we track a small set of indicat