The latest education technology argument is not really about whether schools should use artificial intelligence. It is about whether AI tools are being designed around how people actually learn.
That distinction became sharper this month as parents and education experts raised new concerns about AI use in U.S. classrooms. A June 23, 2026 report in The Guardian described parent campaigns calling for limits or moratoriums on generative AI in schools, including worries that classroom AI assignments can encourage cognitive offloading, shallow work, and overdependence on automated answers. At the same time, school systems and teacher organizations are trying to move from panic to training. The American Federation of Teachers’ National Academy for AI Instruction frames AI as something educators need to shape, not simply accept from vendors.
The practical issue is that “AI tutor” can mean very different things. It can mean a chat box that gives quick answers. It can also mean a carefully constrained learning system that asks students to predict, explain, retrieve, revise, and try again. Learning science strongly favors the second version.
What Happened
The current AI-in-school debate is moving on two tracks. One track is adoption: teachers, districts, and education companies are experimenting with AI lesson planning, feedback tools, tutoring bots, writing support, and personalized practice. The other track is caution: parents and researchers are asking whether students are learning more or simply outsourcing more of the mental work that builds skill.
Recent research gives both sides useful evidence. A 2025 randomized controlled trial published in Scientific Reports found that students using a custom AI tutor in a college physics setting learned more in less time than students in an active-learning classroom condition. But the important detail is easy to miss: the tool was not just a general chatbot. It was built around instructional structure, sequenced problems, and research-based pedagogical design.
A different study, published in PNAS, found the danger on the other side. When high school math students used generative AI without proper guardrails, the tool improved practice-session performance but could reduce later learning when students had to work independently. In other words, AI made some students look better while they were using it, then left them less prepared when the support disappeared.
That contrast is the real news. AI tutoring is not one intervention. It is a design choice. A recent Brookings review of generative AI tutoring research makes a similar point: tutoring benefits depend on whether the tool is built to support sound instruction, not on whether it uses generative AI.
Why It Matters
Learning requires effort that cannot be fully delegated. Students need to retrieve information from memory, notice errors, compare strategies, explain reasoning, and make decisions without constant support. These processes can feel slower than receiving an answer, but they are the work that makes knowledge usable later.
That is why unstructured AI help is risky. If a student asks a bot to solve the problem, summarize the chapter, write the outline, or generate the example before thinking, the student may finish the assignment while avoiding the practice that would have strengthened memory and judgment. The product looks complete, but the learner has done too little of the cognitive work.
Well-designed AI could help with the opposite pattern. It can ask a student to try first, request an explanation, give a hint instead of an answer, diagnose a misconception, vary the next problem, and prompt reflection after an error. That is closer to tutoring because it preserves productive struggle and uses feedback to improve the next attempt.
The difference matters for assessment, too. If schools only ask for take-home products that AI can generate, students will be tempted to optimize for completion. If schools also use in-class retrieval, oral explanation, drafts with feedback, problem-solving checks, and revision notes, students have clearer reasons to use AI as a coach instead of a substitute.
The Practical Learning Conclusion
The useful rule is simple: AI should increase student thinking before it increases student output. If the tool mainly makes work faster, it may be convenient. If it makes the learner retrieve, explain, check, and revise, it has a better chance of supporting durable learning.
- Ask before answering. A useful AI tutor should require the student to attempt the problem, prediction, thesis, or explanation before giving help.
- Prefer hints over solutions. Good support narrows the next step without removing the need to reason through it.
- Build in retrieval. Students should close notes, recall what they know, and then use AI to check gaps or generate targeted practice.
- Require explanation. If a student cannot explain why an answer works, the AI interaction has produced completion, not learning.
- Keep independent checks. Short no-AI quizzes, whiteboard explanations, oral defenses, and handwritten problem solving show whether the skill transfers without the tool.
- Track errors, not just scores. The best AI feedback should help students name the misconception and practice a corrected version.
For teachers and parents, the question should not be “Is AI allowed?” The better question is: “What mental work is the student still required to do?” AI becomes educational when it protects that work. Without that learning-science foundation, the newest EdTech trend can become an answer machine with a tutor’s name.