AI in learning: help or hindrance? Stanford’s findings

A seemingly flawless piece of homework produced with artificial intelligence does not, on its own, show what a student has understood. According to the official research review by Stanford University’s SCALE initiative, handing over the completion of tasks can undermine independent learning and recall, while using AI to support understanding, practice and feedback can help learning. For parents and teachers, it is therefore worth clarifying not only whether students may use such tools, but exactly what they may use them for.

A completed task is not the only measure

The SCALE review examined 52 studies investigating cause-and-effect relationships. The university’s summary makes an important distinction between technology doing the thinking for a learner and helping them do that thinking themselves. The findings should therefore be read neither as a blanket endorsement of AI nor as grounds for rejecting all its educational uses.

Whether an assignment has been completed is one question; whether its author could later explain the reasoning behind it is another. The latter also needs a place in the assessment of learning. A well-worded answer may reflect understanding, but it may simply be borrowed text.

When does help become the outsourcing of thinking?

In this context, cognitive offloading means that the learner hands over the essential thinking involved in a task to the tool. Stanford’s review links this kind of delegation of homework and written assignments to poorer independent learning and recall. This does not mean that every request for an explanation or suggestion for improvement would have the same consequences.

According to the report of a study published in PNAS, students who received direct help with solutions progressed faster during practice but later performed worse on tests taken independently. The lesson should be interpreted narrowly: immediate performance achieved with assistance does not automatically mean that the knowledge will also be available without it. This does not establish that AI has the same effect across all subjects and age groups.

What kind of help better supports learning?

The SCALE summary identifies understanding new concepts, practising and requesting feedback on one’s own work as uses that support learning. What these have in common is that the student still has a meaningful task to carry out. They must explain what they do not understand, attempt a solution and then weigh up the help they receive.

In practice, this might mean asking for feedback on weaknesses in one’s own outline rather than requesting a finished essay, or asking for a hint about the next step rather than a complete mathematical solution. The sample prompts below illustrate this distinction; they are not tested teaching formulas that apply in every situation.

• For understanding: “Explain this concept in a different way, then ask me questions to check my understanding.”
• When stuck: “Do not give me the solution; just point out which of my steps I should reconsider.”
• For written work: “Point out where my text needs supporting reasoning, but do not rewrite it for me.”
• For practice: “Give me a similar task, and let us discuss the solution only after I have attempted it myself.”

A framework for home: independent effort, targeted help and independent checking

Establishing a simple routine can be a useful starting point for parents. The student first makes an independent attempt, then identifies where they are stuck and asks for help only with that part. This shifts the conversation away from finishing homework as quickly as possible and towards identifying what still needs to be understood.

After receiving help, it may be worth setting the conversation aside and asking the learner to explain the steps of the solution in their own words. Another option is to solve a similar task independently. These approaches do not replace assessment by a teacher, but they give families an opportunity to discuss learning without relying solely on the finished answer.

Parents do not necessarily need detailed expertise in every subject to do this. They might ask, for example, “Where did you get stuck?”, “Why did you choose this solution?” or “What did you change after receiving feedback?” The aim is not to introduce another oral test at home, but to encourage students to articulate their own reasoning.

Setting assignments: make the reasoning visible

For teachers, it may be useful to distinguish in advance between work for which assistance is allowed and work that must be completed independently. Asking for an explanation of a concept, requesting feedback on one’s own text and having an entire assignment written are different activities. Giving specific examples in the task instructions may therefore be more useful than simply stating that AI use is permitted or prohibited.

One possible approach is to ask students to submit an outline, a first draft and an explanation of their revisions together. Students can briefly describe what assistance they used, which suggestions they accepted and which they rejected. This is an editorial and pedagogical suggestion, not a method that the review establishes as effective in every classroom.

Assessment could also include a short oral explanation or a smaller task to be completed independently. This makes it possible to discuss the learner’s decisions and understanding alongside the finished work. The emphasis need not be on detecting attempts to get around a ban, but on making the learning objective clear.

What cannot yet be stated in general terms?

According to Education Week’s report, there are still few studies conducted directly in US primary and secondary school classrooms. The findings should therefore be applied cautiously to other age groups, subjects and educational settings. An advantage or disadvantage observed in a research setting does not automatically predict what will happen in a Hungarian classroom.

Nevertheless, the practical lesson from Stanford’s review offers a clear direction: it is worth distinguishing uses that support understanding, practice and feedback from handing over a task entirely. The shared measure should not simply be how polished an answer looks, but also what the student can explain and apply independently. This gives AI a clearly defined role in helping to achieve the learning objective, rather than allowing it to take that objective’s place.