Artificial intelligence has entered the classroom faster than school policies have been able to keep up. Students of all ages now use chatbots to outline essays, solve algebra problems, check their written work, and retrieve answers before teachers have explained the ideas behind them. The early evidence paints a complicated picture: AI can deliver immediate improvements in assignment quality and performance, but it does not always leave students with a deeper understanding.

That gap between performance and learning lies behind a major policy shift. New York City, the largest public school district in the United States, has introduced a one-year moratorium on student-facing generative AI for public school students through the eighth grade. The measure is one of the toughest responses so far to the spread of AI in classrooms. It is also, in the view of many parents and researchers, a response that has come late and may not go far enough.

The scale of AI adoption among young students helps explain the concern. A 2026 Pew Research survey found that 54 percent of students had used AI to help with schoolwork, and 12 percent said they relied on it for every assignment. In a 2025 report from the Center for Democracy and Technology, 85 percent of teachers and 86 percent of students reported using AI during the 2024-25 school year. These numbers suggest that AI has become a standard educational tool, often without clear guidance about what it should and should not do.

The double-edged nature of AI tutoring

Several recent studies illustrate why the simple question, does AI help students perform better?, receives an easy yes, and the more important question, does AI help them learn?, receives a far more cautious no. Researchers at Stanford University followed K-12 students working on math, writing, and coding tasks. With AI assistance, many students completed assignments more quickly and performed at a higher level than their unaided peers. When the AI tools were removed, however, most of those advantages disappeared. Some students, especially those comfortable with generic chatbots, had come to depend on the technology so heavily that their unaided performance was worse than it had been before.

The Stanford findings align with a broader body of cognitive science research. For example, struggle and retrieval are essential to memory formation. A student who asks an AI for an instant answer avoids the mental friction that helps concepts move from short-term working memory into long-term understanding. The learner might deliver a flawless paragraph or solve a problem without making the conceptual leap that the assignment was designed to produce. The tool has done the thinking, and the student has practiced the role of editor, or perhaps only that of orderer.

There is an important distinction in the Stanford work between general-purpose AI assistants and learning-specific tools. Well-designed tutoring systems can provide prompts, counterquestions, and hints that preserve the student's cognitive work. Generic assistants, by contrast, tend to simply complete the requested task. The former can support a healthy learning loop; the latter, especially when used without guidance, can become a shortcut that undermines autonomy.

Another study from the Brookings Institution reached a similar conclusion. It found that well-designed AI tools, used as part of an overall and pedagogically sound approach, can benefit classrooms, increasing engagement and freeing teachers to give more individualized attention. But an overreliance on AI chatbots, the researchers warned, can harm a student's ability to reason independently and

Source:PCWorld News