Artificial intelligence presents education with a peculiar problem. We have invented a machine capable of doing many of the intellectual tasks that schools are supposed to teach children how to do. It can summarise the chapter, construct the argument, solve the problem, improve the prose, suggest the hypothesis and sometimes write the entire assignment. The temptation is obvious. Why struggle for forty minutes when a chatbot can produce something respectable in forty seconds?
The Australian Education Department is now warning that there may be a substantial price for that convenience. In evidence to the parliamentary inquiry into artificial intelligence, it warned that frequent reliance upon AI can weaken critical thinking and the "thinking infrastructure" required for future learning. The danger has a name: cognitive offloading. Instead of using the mind to perform a difficult operation, we hand the operation to the machine.
There is nothing inherently wrong with cognitive offloading. Human civilisation is built upon it. We write things down rather than memorising everything. Calculators save us from tedious arithmetic. Maps save us from memorising every road. Books preserve knowledge outside the human brain. Nobody seriously proposes throwing away these technologies.
But there is a fundamental difference between offloading a capacity one possesses and never developing that capacity in the first place.
Give a calculator to somebody who understands arithmetic and it is an extraordinarily useful tool. Give it to a child before the child understands what multiplication means and the calculator can become a substitute for mathematical understanding. The answer appears on the screen, but the mental machinery that makes the answer intelligible may never be properly constructed.
Generative AI extends that problem far beyond arithmetic. A child can now avoid the intellectual friction involved in almost every school subject. Faced with a difficult paragraph, ask AI to summarise it. Faced with an essay, ask AI for the argument. Faced with competing interpretations, ask which is correct. Faced with a blank page, ask it to write the first paragraph. Faced with an awkward sentence, have it rewritten. Each individual shortcut appears trivial. Accumulated across thousands of school tasks, however, the child may gradually outsource precisely the activities through which intellectual competence develops.
The latest Australian figures are certainly concerning. Analysis of the 2025 PISA results found that Australian students who reported using AI for schoolwork weekly or more scored around 30 points below infrequent users in science and 32 points below them in reading, after adjustment for socioeconomic background. The difference in mathematics was smaller, at 14 points.
That does not prove that AI caused the lower scores. Perhaps weaker students are more likely to reach for AI in the first place. Perhaps both processes operate together. The sensible conclusion is therefore not that a chatbot has been scientifically demonstrated to make a child stupid. It is that substantial AI dependence is associated with poorer educational performance, while there are good cognitive reasons for worrying about excessive dependence.
Those reasons become clearer when we remember what learning actually involves.
Struggle is not an unfortunate side effect of education. Much of the time, struggle is education. Trying to remember something strengthens memory. Trying to formulate an argument forces the mind to organise information. Writing a bad first paragraph and improving it teaches composition. Getting a mathematical problem wrong and discovering why develops mathematical understanding. Reading a difficult page twice develops concentration. AI can remove all of these inconveniences. Unfortunately, it can remove the learning occurring inside them as well.
This is particularly important for children because they have not yet accumulated the intellectual resources that adults bring to AI. An experienced lawyer can ask an AI system to summarise a judgment and recognise that an important qualification has disappeared. A scientist can use it to explore a hypothesis while noticing when the machine produces scientific nonsense. An experienced writer can ask for criticism and reject bad stylistic advice.
A twelve-year-old cannot necessarily do any of those things. To evaluate the machine, one must possess knowledge independent of the machine. Otherwise the child encounters a polished paragraph on a screen and mistakes fluency for truth.
That is the paradox of teaching "AI literacy." The most important preparation for using artificial intelligence may be developing intelligence without it. Children need vocabulary before automated writing. They need arithmetic before calculators. They need factual knowledge before automated research. They need to write arguments before asking a machine to improve arguments. Most importantly, they need experience being wrong, thinking again and discovering why they were wrong. Critical thinking cannot exist in a vacuum. One cannot critically evaluate information about which one knows nothing.
This does not require banning AI from education. That would probably be both impractical and educationally foolish. Children will enter universities and workplaces in which AI is commonplace. Schools that pretend otherwise would be preparing students for a world that has already disappeared.
The answer is sequencing. In the early years, foundational intellectual work should remain overwhelmingly human. Children should read books, write by hand, learn spelling and grammar, memorise essential facts, practise arithmetic, solve problems and explain their reasoning without an electronic oracle continually sitting beside them. AI should not become the intellectual equivalent of the parent who completes the child's homework because watching the child struggle is inconvenient.
As students mature, AI can gradually become an object of criticism and then a tool. Instead of saying, "Ask the chatbot to write an essay about Federation," the teacher might require the student to write the argument first and then ask AI to challenge it. Which objections are good? Which are bad? What did the machine overlook? Can the student find a factual error? Can two different AI systems be prompted to argue opposite sides and can the student decide which argument is stronger?
That reverses the relationship. The student is no longer the machine's apprentice. The machine becomes material upon which the student's intelligence operates.
There is experimental evidence that such scaffolding matters. A recent study of sixth-grade students found that structured learner-peer-AI interaction designed specifically to prevent cognitive offloading improved critical-thinking measures, particularly questioning and verification. This points towards a much more sensible educational model than either unrestricted chatbot use or prohibition.
Assessment must change as well. If schools continue assigning work that a chatbot can complete from beginning to end, students will inevitably use chatbots to complete it. More work will need to occur in classrooms. Oral examinations deserve revival. Students can be asked to defend an essay rather than merely submit it. Teachers can require drafts, notes and explanations of how conclusions were reached. Handwritten work has a place precisely because education is interested in the process occurring inside the student's head, not merely the polished product eventually handed to the teacher.
Parents have a role too. Asking AI a factual question is one thing. Having it do homework is another. A useful household rule would be simple: your brain goes first. Read the material. Attempt the problem. Write the paragraph. Form an opinion. Only then use AI, and preferably use it to test what you have done rather than replace it.
There are additional reasons for caution. Australia's eSafety Commissioner found in a 2026 survey of 1,950 children aged 10 to 17 that 78 percent had used an AI assistant. Among children who had used AI assistants or companions, 20 percent used them daily or more frequently. Almost a third had disclosed personal or potentially sensitive information, while one in five reported potentially inappropriate or harmful interactions. AI is therefore no longer merely an educational technology. For some children it is becoming part tutor, part search engine, part adviser and even part companion. That makes intellectual independence more important, not less.
UNESCO's guidance makes essentially the same point from another direction. Literacy, numeracy and scientific knowledge remain foundational even where AI can outperform humans at particular tasks. Students need to learn to evaluate AI-generated material critically rather than merely becoming efficient consumers of it.
The great educational mistake of the AI age would be to confuse producing an answer with knowing something. A child who asks AI to write a beautiful essay may possess a beautiful essay. That does not mean the child can write. A child who receives the solution to a mathematical problem possesses the solution. That does not mean the child understands mathematics. A student who receives an instant summary of Macbeth possesses information about Macbeth. That does not mean the student has read Shakespeare, wrestled with the language or developed the concentration required to read another difficult book.
Artificial intelligence may eventually become one of the greatest educational technologies ever invented. A patient AI tutor that explains something ten different ways, never becomes irritated and adapts instruction to an individual child has obvious possibilities. But a tutor is supposed to make the pupil think.
If AI instead becomes the thing that thinks so the pupil does not have to, we will have achieved something technologically remarkable and educationally disastrous: machines of unprecedented intellectual power accompanied by humans increasingly unable to function without them.
The principle for schools and parents should therefore be simple. Do not try to raise children in a world without artificial intelligence. That world is gone. Raise children whose own intelligence is strong enough to use it.