Schools Are Rushing Into AI: Here’s What They’re Missing

Artificial intelligence (AI) in education is moving beyond early enthusiasm for tools, automation and personalization toward a deeper debate about pedagogy, ethics, literacy and human agency, according to a new systematic review.

The review, published in Information, examined 235 international studies on artificial intelligence and pedagogy published between 2005 and 2025. It warns that visible adoption does not automatically produce better learning. AI can personalize instruction, support feedback and expand access to resources, but it can also amplify inequality, blur authorship, encourage dependence, weaken assessment and shift control toward private platforms.

From smart tools to difficult questions

Large language models (LLMs) and chatbots do not merely help students find information. They produce explanations, essays, summaries, code and arguments that can look credible even when they are flawed, making them different from earlier digital learning tools. They do not simply assist learning; they intervene in the production of knowledge itself.

The review finds that recent research has become more concerned with academic integrity, authorship, transparency, cognitive dependence, assessment redesign and teacher preparation. The shift shows that education systems are beginning to understand that the biggest AI challenge is not cheating alone. Cheating is only the most visible symptom of a deeper disruption.

If students can use AI to generate work, educators need to rethink what counts as evidence of learning. If AI can produce plausible explanations, students need to learn how to test, question and verify them. If teachers use AI to grade or give feedback, institutions need rules on privacy, fairness, accountability and professional judgment.

Banning tools may be tempting, but it is unlikely to be enough. The study points toward a more difficult task: redesigning learning so that students develop the ability to work critically with AI rather than simply outsource thinking to it.

Four lenses for understanding the AI classroom

The review identifies four major ways researchers now understand AI in education: critical, ethical, literacy-oriented and humanistic.

• The critical lens asks who benefits from AI and who may be left behind. It treats AI not as a neutral innovation, but as a technology shaped by power, infrastructure, language, data and access. This is especially important for developing countries and low-resource education systems. AI may promise personalized learning, but if students lack connectivity, teachers lack training, or tools are not available in local languages, the result may be deeper inequality.
• The ethical lens focuses on responsibility. It includes academic integrity, privacy, transparency, data governance and institutional safeguards. The review makes clear that ethics cannot be reduced to telling students not to misuse AI. Institutions also need clear rules for how AI tools are selected, how student data is protected, how automated judgments are reviewed and how teachers are supported.
• The literacy-oriented lens is the broadest. It argues that AI literacy is not simply knowing how to use a chatbot. Students and teachers need to understand how AI systems work, what kinds of errors they produce, where bias can enter, and when human judgment must override machine output. In this sense, AI literacy is becoming a core educational competence, not a specialist skill reserved for computer science.
• The humanistic lens asks what happens to creativity, agency, dialogue and identity when AI enters the learning process. Education is not only about efficiency. It is about forming judgment, curiosity, empathy, creativity and civic capacity. AI can support those goals only if it is designed and used around human development, not merely automation.

Together, they offer a practical framework for policymakers and education leaders.

The Global South cannot be an afterthought

Because the study relies on Web of Science and Scopus, it captures internationally visible academic research, but may underrepresent regional databases, multilingual scholarship, community-based work and local experiences from the Global South.

Many AI education frameworks are written from the perspective of well-resourced institutions. They assume access to devices, connectivity, teacher support, digital infrastructure, English-language tools and regulatory capacity. Those assumptions do not hold everywhere. In many education systems, the first-order barriers are not advanced AI governance models but electricity, bandwidth, teacher shortages, language diversity and unequal access to basic digital tools.

For developing countries, the lesson is to avoid importing AI policy as a generic package. Education ministries and development agencies need locally grounded strategies. That means investing in teacher training, public digital infrastructure, local-language tools, inclusive design and evidence on actual learning outcomes.

The risk is that AI becomes another layer of educational inequality: advanced tools for wealthy systems, fragile experimentation for poorer ones. The opportunity is that, with careful design, AI could help expand access to learning support, assist teachers, improve feedback and open new pathways for students who are underserved by current systems.

Better Learning, Not Better Technology

AI adoption is not the same as educational transformation, the study asserts. The next phase should be less about tool announcements and more about hard implementation questions. Are teachers trained to use AI critically? Are students learning to evaluate AI-generated content? Are assessments being redesigned to measure reasoning, creativity and applied understanding? Are data and privacy protections clear? Are low-resource schools being included? Are local languages and cultural contexts represented? Is there evidence that AI improves learning outcomes?

The future classroom will almost certainly include artificial intelligence, but whether that classroom becomes more inclusive, thoughtful and effective will depend less on the power of the tools and more on the choices institutions make around them.