AI & Education

Artificial intelligence is raising important questions about how we teach, learn, and think. What role should it play in education? When can it support understanding and creativity—and when might it take over the thinking learners need to do themselves?

This section helps educators, families, and curious readers explore AI with informed curiosity and a critical eye. We’ll examine classroom uses, lesson planning, feedback, accessibility, and student learning, alongside concerns about accuracy, bias, privacy, academic honesty, and originality. We’ll also explore how to evaluate AI-generated information and help learners use these tools responsibly.

Through research, practical examples, and thoughtful discussion, our purpose is to help readers make informed choices about AI’s place in education—keeping independent thinking, meaningful learning, and the relationships between teachers and students at the center.

When Does AI Support Learning—and When Does It Do the Thinking for Us?

A student submits a polished paragraph. Another finishes a difficult math assignment. A third arrives with a neatly organized summary of a chapter. Each has used artificial intelligence. The work looks complete—but what has each student learned?

That question belongs at the center of conversations about AI in education. Finishing an assignment and developing understanding are related, but they are not interchangeable. To judge whether AI is helping, we need to look beyond the finished product to the thinking the learner actually does.

What does generative AI do?

Generative AI produces content, including text and images, in response to instructions. Text-generating systems use patterns learned during training to construct responses. They can offer explanations, examples, questions, and drafts, but fluent wording is no guarantee of accuracy. A convincing response can contain mistakes or invented references.

For education, this means an AI response should be something to examine, not automatically accept. Students need opportunities to compare explanations with reliable sources, notice inconsistencies, and ask whether an answer makes sense.

What research can tell us

A 2025 study by Hamsa Bastani and colleagues, published in Proceedings of the National Academy of Sciences, examined nearly 1,000 high school mathematics students in Turkey. Researchers compared practice with no AI, a general-purpose GPT-4 interface, and a specially designed GPT-4 tutor with safeguards intended to guide learning.

Both AI groups performed better during assisted practice. However, on a subsequent exam without those resources, the general-purpose AI group scored about 17 percent lower than the no-AI group. This was a relative reduction, not a 17-percentage-point drop. The tutor group largely avoided that negative effect, but did not show a statistically significant advantage over the no-AI group on the independent exam.

This was one study in a particular setting, not proof that every AI tool helps or harms every learner. Its central lesson is nevertheless important: stronger performance while using a tool does not necessarily mean stronger learning afterward. [1]

Support or substitution? Start with the learning goal

Before deciding whether AI belongs in an activity, ask: What is the student supposed to learn here?

If the goal is constructing an argument, having AI supply the argument may bypass the work the student needs to practice. If the goal is evaluating arguments, examining an AI-generated example could be a useful starting point. The same technology can serve very different purposes.

The following examples are practical suggestions, rather than activities tested in the study above:

Learning taskA possible supporting roleWhen AI may take over the intended thinkingSolving a math problemOffer one hint, then ask the learner to explain the next step.Supply a complete solution for the student to copy.Developing a paragraphAsk questions about unclear ideas in a student-written draft.Write the paragraph before the student develops an idea.Understanding a readingAsk questions after reading and have the student locate supporting evidence.Provide a summary that replaces reading when reading and interpretation are the goals.Preparing for an assessmentGenerate questions for the learner to answer, with answers checked against course materials.Provide answers the learner reviews without trying to retrieve or explain anything.

Even a hint can give away too much, and feedback can be wrong. These approaches still require judgment and oversight. A worked example can also be useful instruction when students examine its reasoning and then attempt a new problem themselves.

The teacher's role: keep the learning visible

Teachers can make the boundaries clear before students begin: what assistance is allowed, what students should produce themselves, and how they should acknowledge AI use.

For an approved, age-appropriate tool, a useful classroom routine might be: attempt the task, seek limited help, explain the reasoning, and try a similar task without AI assistance. The final step gives the teacher evidence about what the student understands beyond the supported activity.

AI may also help a teacher draft practice questions, suggest examples, or prepare different explanations. Those materials need review for accuracy, clarity, and fit with the lesson. A resource that looks polished still needs the teacher's knowledge of the subject and the students.

UNESCO's guidance calls for a human-centered approach that considers educational purpose, privacy, age appropriateness, and human agency. [2] In practice, that means using approved tools, protecting student information, and keeping educators responsible for instructional decisions. Younger learners can encounter teacher-selected examples without independently using a chatbot.

Support should also protect access

Independent thinking does not require every learner to work without accommodations. A student who uses speech-to-text may still generate every idea in a paragraph. A student who hears a text read aloud may still interpret it and support an argument with evidence. These tools are not necessarily generative AI, but they illustrate an important distinction.

We need to ask whether assistance removes an access barrier or replaces the particular skill being taught. Those decisions depend on the learner and the purpose of the activity. When checking understanding without AI, students should retain the accommodations they need.

Questions families and students can ask

Families do not need to become technology experts to begin a useful conversation. After an AI-supported task, ask:

  • What part did you do, and what part did the tool do?

  • Can you explain the answer in your own words?

  • How did you check the information?

  • What did you change or disagree with, and why?

  • Could you try a similar task without the AI's help?

These questions invite reflection rather than merely checking whether homework is finished. They also make it easier to notice when a learner needs more explanation, practice, or direct teaching.

AI's educational value depends on what happens in the learner, not simply what appears on the screen. The most useful question to carry forward is: After this help, what does the student understand or know how to do?

Something to consider: Think of a task a student completes regularly. Where could AI help the student learn—and where would its involvement undermine the purpose of the task?

References and further reading

  1. Bastani, H., et al. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122. https://doi.org/10.1073/pnas.2422633122

  2. Miao, F., & Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research