UNESCO: Putting Human Agency at the Heart of AI Education

A senior UNESCO official leading the Organization's work on technology and artificial intelligence in education has warned that the rapid spread of AI across universities and schools is producing what researchers have termed "cognitive surrender", and has urged institutions worldwide to build the strategies and evidence base needed to reverse it.

Addressing an international gathering of educators, university leaders and industry partners convened by Huawei, the official said AI is now embedded in the daily lives of growing numbers of students and teachers. The emerging evidence, however, points in a troubling direction.

"More and more use of AI has not translated into better learning and better thinking," the official told delegates. "Most of us who are using AI are offloading our thinking to the machine. We are surrendering our writing, we are surrendering our thinking."

The term "cognitive surrender" was introduced in this context by Steven Shaw and Gideon Nave of the Wharton School in a preregistered study released as a working paper in January 2026. They define this phenomenon as the uncritical acceptance of AI outputs in place of one's own reasoning. Across three experiments with 1,372 participants, the researchers embedded an AI assistant that randomly returned either correct answers or confidently delivered wrong ones. In the study's central experiment, participants accepted the AI's faulty answers roughly 80 percent of the time and reported greater confidence in their responses than those who worked without AI. The finding sharpens a longer-standing research literature on cognitive offloading, which involves the delegation of mental work to external aids, by identifying the point at which delegation becomes dependence: where little or no independent view is formed against which the machine's answer can be checked.

This concern is now reflected in the Preliminary Report of the Independent International Scientific Panel on AI, released on 1 July 2026, that was presented to Member States at the first Global Dialogue on AI Governance in Geneva this week. The Panel, established by the United Nations General Assembly and co-chaired by Yoshua Bengio and Maria Ressa, cautions that cognitive offloading, in which users delegate mental work to AI rather than using it to support their own reasoning, can weaken critical thinking. In education, the Panel identifies an "illusion of competence": students may appear to perform better while using AI without developing the underlying thinking capabilities needed to repeat the work independently.

The Panel also suggests that this effect can be mitigated. One strategy would be for AI systems to be designed with safeguards that prompt students through the reasoning process, instead of supplying complete answers. Gains from AI in education were strongest when systems were built for clearly defined tasks, embedded in human-centered workflows and supported by AI capable teachers and peers. For UNESCO, these prospects for human-centred design have to be matched with cultivating the critical agency among teachers and students in their use of AI for learning.

From tool to agent

Speaking as a UNESCO representative but also as a teacher and professor with long experience of working closely with students, the official framed the challenge as one of the fundamental pedagogical questions of the age of generative and increasingly agentic AI: how learners and teachers can reclaim human cognition and human agency.

AI, the official observed, is no longer just a tool. It is increasingly an agent, capable of acting with growing autonomy outside human control. UNESCO's response has been to publish AI competency frameworks for teachers and for students, anchored in commitments to human rights, human agency, equity, inclusion and ethics. Both frameworks prioritize strategies to cultivate agency, ethical judgement, foundational AI knowledge and a creator's mindset: the disposition to shape the technology instead of consuming it uncritically. Students are encouraged not to accept ready-made answers or submit polished essays that the machine has in effect written, but to question, reason, and argue through critical exchange with AI and with their peers. For teachers, the framework calls for a corresponding shift toward assessing the learning process, not just its outputs.

Evidence from the classroom

The official pointed to the practice of Huawei's Most Valued Instructors as a demonstration of these practices in action. What the competency frameworks describe in near-theoretical terms, these instructors are enacting daily: keeping students engaged in cognitive struggle and cognitive friction, requiring them to muddle through, apply their executive function, use AI to scaffold reasoning, and learn by arguing, applying and building prototypes collaboratively and importantly, evaluating critically, their application in real world settings.

This convergence between frontline practice and the Scientific Panel's evidence on safeguarded, reasoning-focused AI use was, the official suggested, the most significant lesson of the event. Yet these instructors represent a small fraction of the education community, working within institutional systems and curricula that have not caught up with them.

Three calls to action

The official closed with three calls to action, each sharpened by the Panel's central caution that policymakers need scientific evidence to govern AI effectively, but that by the time the evidence is clear, it may be too late to act.

First, build the evidence base. The case studies presented at the event demonstrate pedagogical leadership but remain anecdotal. Systematic documentation is needed so that organizations working globally on institutional and system change can draw on this knowledge with confidence.

Second, move critical pedagogical practice from a few isolated departments in a given institution towards system-wide adoption. The educators leading this work must not remain an exceptional minority; their cases of critical pedagogical practice should be leveraged to catalyze change across institutions and systems, including those that are under-resourced and lack the human capacity to navigate AI critically.

Third, partner on shared infrastructure such as practice labs, turning individual, small-scale innovation into systemic change. Infrastructure of this kind would generate evidence and institutional change simultaneously, a direct response to the Panel s warning that waiting for conclusive evidence can also mean acting too late.

Soften or sharpen

"We have to decide as a people whether we will use AI to soften the human mind or to sharpen it," the official concluded. That decision runs in two directions at once. One is cultivating the human capabilities to engage critically with AI, the work the competency frameworks set out for teachers and students. The other is building technologies designed to strengthen human agency. The Scientific Panel's evidence shows this second choice is real and already consequential.