Before the Evidence Arrives: Should Schools Slow Down on Student-facing AI? |

BY BINDU VEMIREDDY*

Education systems around the world are exploring how artificial intelligence might fit into student learning, even as evidence about how to use it well is still emerging. A growing number of researchers and policy analysts now argue that the most sensible response would be to slow down.

A Case for Caution

The latest weight behind that view comes from the Brookings Institution. In a report titled “A New Direction for Students in AI World”, its Center for Universal Education collected inputs from over 500 participants across 50 countries, along with a review of more than 400 articles. The report concludes, given current patterns of use, the risks of generative AI for students outweigh its potential benefits.

In a follow-up commentary, Brookings framed its position as “go slow to go fast.” It says that policymakers and educators should bring families into the conversation and make sure the technology serves children’s learning and development before student-facing AI expands at scale.

The concern is not simply that students might cheat. It is that some of the effects may be invisible at first. Many study participants worried that routine overuse of AI could lead to cognitive decline, much as muscles weaken without use.

A Lesson from Estonia

One of the clearest early signals comes from Estonia, a country known for digital governance. Its AI Leap program in secondary schools paired AI tools with teacher training and AI literacy instruction. Even so, many students reported turning to AI before approaching teachers with questions, leading to fewer student-teacher interactions.

The finding suggests that even a carefully designed rollout can quietly change how students relate to their teachers, a relationship that is central to learning.

What Education Is For

The debate lands on an old question about what schooling is meant to produce. In 1947, Dr. Martin Luther King Jr. wrote in “The Purpose of Education” that education must enable a person “to sift and weigh evidence, to discern the true from the false.” Decades later, historian William Cronon listed among the aims of a liberal education the ability to “break it into pieces, and figure out how it works,” so that people can act sensibly in the real world.

Neither writer was thinking about AI. Both described capacities that come from years of effortful practice: recalling facts, planning an argument, reasoning through a problem, drafting and redrafting. These are also tasks that generative AI can help students complete, raising questions about how much of the underlying thinking students are doing themselves.

The Offloading Problem

Researchers call this cognitive offloading, and it is drawing serious attention. The Education Endowment Foundation is examining how handing recall, planning, reasoning and drafting to AI may affect knowledge acquisition, conceptual understanding, working memory and problem-solving. It is also asking whether the effects differ for students from socioeconomically disadvantaged backgrounds. Findings are still to come.

Brookings’ researchers describe a mechanism that could make offloading harder to resist. Generative AI has “turbocharged” offloading, especially in schools where learning can feel transactional. If a task can be completed without reading, discussing or reflecting, the tool becomes a shortcut around the very thinking the task was designed to build.

The Other Side of the Argument

The caution is contested. Critics of restrictions on student-facing AI argue that limiting access will leave young people unprepared for an AI-enabled future. The fear of being left behind is understandable, and it is widely shared by parents and policymakers.

The Brookings report has its own critics. A review published by the National Education Policy Center found it useful for anticipating the pros and cons but weaker on a clear blueprint for change. Others stress that the report is not a call to abandon the technology. They argue that AI is more likely to help when it is introduced after students have engaged with the material and can explain their reasoning.

Sequence Over Speed

That test points to a way of reframing the debate. The strongest argument for caution is not that AI is harmful. It is a question of order. Adults who use AI have already built the judgment to question its output. Children are still building it.

The worry that students will fall behind if they do not start early may therefore be only half the picture. The other half is what happens when students begin outsourcing parts of their thinking before the underlying skills exist. Critical thinking does not appear later, it must be built first, and AI has made it easier than ever to skip part of that process.

An Open Question

Policymakers do not need to choose between embracing AI and banning it. They do need to decide how much evidence to require before expanding student-facing tools, and who bears the risk if that evidence turns out to be unfavorable. For families and schools, navigating these choices abroad, the question is practical and immediate.

Are schools giving students enough room to develop critical thinking, or moving faster than the evidence supports? On the current research, the honest answer is that no one yet knows. That uncertainty is itself a reason to go slow to go fast.

Sources & Further Reading:

*Bindu Vemireddy is a technology entrepreneur, strategy and operations leader, and early childhood education advocate based in metro Atlanta. She is the co-founder of Assistly, an AI-enabled platform for senior care providers, and Chief Strategy Officer at Orcaworks, where she focuses on responsible AI adoption and enterprise automation. For more than a decade, she has also owned and operated two Goddard Schools in metro Atlanta, reflecting her commitment to early childhood education. She supports initiatives that expand access to education and financial opportunities in India.