OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education

Both non-generative AI and GenAI can be used to automate some educational tasks performed by teachers, students, administrators or other stakeholders – even though it cannot lead to an automation of educational occupationq as a whole in the foreseeable future. Human judgment, relationships or responsibility should not be inadvertently delegated to GenAI tools, even though some educational tasks could purposefully be automated. The OECD-EI Guidelines highlighted the importance of keeping a “human in the loop” and, when possible, a human alternative to digital tools or processes. This is key to implementing the human-centred values of the OECD AI Principles. The above insights already include this focus on human agency. This section concludes the document by emphasising this key dimension.

As mentioned above, GenAI has the potential to automate some time-consuming teaching, learning, assessment and administrative tasks and, in principle, allow teachers and education staffto redirect their efforts to more important learning, teaching or administrative tasks, including relational ones. Still, in most cases, the human agency and oversight of students, teachers and administrators when using GenAI is essential. Students need to be agents of their learning for actual learning to happen. Teachers also need to be agents of their teaching to continue to learn professionally and engage students in learning – and avoid skill atrophy as a result of uncritically implementing content generated by GenAI tools. Institutions might consider annual “automation and agency” audits that would lead actors to reflect on what tasks or subtasks they have automated and how they exercise their agency. Developers should ensure that teachers remain active by designing GenAI tools in ways that maintain this agency when necessary, for example by making them co-design some GenAI tools, allow them to customise the AI tools as they wish, or to validate the outputs generated by the AI tools. While human agency can imply to revise or discard GenAI outputs, it also has to be supported by specific design features of GenAI tools. Education researchers should also use GenAI with judgment, imagination and critical thinking, as is the case for all scientists.

The importance of human agency increases when stakes get higher for those affected by the actions or decisions of the GenAI tools. The versatility of GenAI tools may lead humans to become over-confident in GenAI accuracy and overreliant on it. Even when they perform extremely well most of the time, GenAI tools may have some level of hallucinations, make mistakes, lack consistency over time – or just not have the expected effects as human actions because of the relational dimension of education. As a result, humans must exercise their agency, double check and critically evaluate their outputs, especially if social stakes are attached to them.

A dialogue on what should and should not be automated is therefore needed. While human agency is key, its implementation is a distributed responsibility across all AI actors and stakeholders of the education eco-system: governments and jurisdictions, education institutions, students and teachers, as well as companies developing GenAI tools. Keeping human agency at the centre of GenAI use, together with thoughtful decisions about the appropriate degree of automation, can support more educationally meaningful, effective, equitable and trustworthy uses of GenAI in education.