Jonathan Craner
Jonathan Craner

AI can produce an answer in seconds. It cannot produce understanding. Those are two different things, and the gap between them is where the future of education will be decided.

The evidence is already here. RAND surveyed 1,214 American students in December 2025. It found that 62% had used AI for homework, up from 48% just seven months earlier. And 67% of those students said they believed AI use was harming their critical thinking skills. That number rose more than ten points in ten months. The students themselves are telling us they are worried.

That is the tension I want to name. Students are using AI more and trusting it less. They are not confused about the technology. They are confused about what it is doing to them.

The risk is that we automate the wrong part of learning. Gallup reports that 59% of US K-12 educators used AI during the 2025-2026 school year. Nearly two-thirds of those said it freed up time for the parts of teaching that matter most. That is a good thing. But if teachers use AI to generate lessons and students use AI to complete them, the result is a loop. AI teaches AI. Everyone gets an A. Nobody learns.

The problem is not that teachers are using AI. It is that they are doing so without guidance. Gallup found that only 18% of teachers receive formal guidance on how AI should be used at work. For one-on-one instruction and tutoring, 69% receive no guidance at all. Teachers are being asked to navigate a technology that is reshaping their profession, and most are doing it alone.

Education is not one thing. It is four: learning facts and concepts, applying them, creating value with them, and experiencing them in the world. AI can help students build the foundation they need to think critically, but it cannot replace the teacher's role in helping them apply knowledge, question assumptions, and engage with the world. The goal is not to confine AI to one stage of learning, but to ensure it supports each stage without doing the thinking for the student.

This is not an argument against AI. It is an argument for using it to deepen learning rather than bypass it. The Socratic approach, where students are guided through questions instead of simply handed answers, encourages them to explain their reasoning and challenge their own understanding. The flipped-classroom model applies a similar principle: students learn foundational material before class, leaving teachers to use classroom time for discussion, critical thinking, and practical application. Research from RAND points in this direction, suggesting that schools may need to separate AI-assisted learning from classroom practice to prevent students from outsourcing their thinking.

My own approach builds on this distinction through a model I call PSL: Problem, Solve, Learn. AI can help a student solve a problem, but the learning happens when they examine the solution, check the reasoning, identify mistakes, and understand how the answer was reached. The goal is not to remove the challenge of learning, but to use AI to guide students through it. Technology should help students become better thinkers, not simply faster at finding answers.

The challenge of making personalised learning accessible at scale is not new. In 1984, educational psychologist Benjamin Bloom explored the benefits of one-to-one tutoring in his work on the "Two Sigma Problem." Although the size of the reported effect has since been questioned, the underlying challenge remains: how can more students receive individual attention? One teacher cannot sit with thirty students at once. AI could help bridge that gap by guiding students through questions, encouraging them to explain their reasoning and helping them recognise what they do not yet understand. But that only works if the technology is designed to support learning, rather than simply deliver answers.

That is the point I care about most. Nobody has solved AI in education. The technology has been widely used for only a few years, and its long-term effects remain uncertain. No single educator, researcher, or developer has all the answers. We need educators, parents, administrators, researchers, and developers to share what works, acknowledge what fails, and keep testing better approaches. The future of AI in education should not be decided by how quickly students can produce an answer, but by whether they can explain it, question it, and think for themselves. Getting that right is a responsibility we all share.

About The Author

Jonathan Craner is an educator, researcher, and the founder of LearnAI, a proactive AI tutoring platform built on Socratic questioning and Bloom's Two Sigma research. He is also an adjunct professor and conducts research on AI in education with fellow PhD researchers.