AI 进入学校，究竟是好是坏？So Is AI Good or Bad for Schools?

Put the same question about classroom AI to a teacher and to a student, and the two replies will differ — and both will be defensible. That gap is not muddle; it mirrors a tension the literature has not resolved. So rather than declare a winner, this article sets out what each camp actually observes, where the evidence stands today, and which conditions decide whether the technology helps or hurts in a particular class.

The question comes up endlessly, and almost always in a form that implies a single word will settle it. No such word exists. Pretending otherwise flattens a problem with real depth. The place to begin is an uncomfortable fact: careful people who have read exactly the same studies still end up in different camps, each with arguments worth taking seriously. Working out how that happens teaches you more than choosing a team ever could.

Nor is education the first arena where AI's wider influence on thinking and making has drawn this kind of examination. A closely parallel argument — individual gains sitting beside collective unease — plays out in our piece on whether AI dulls our creative instincts.

01The Short Answer

Whether AI helps or harms comes down overwhelmingly to deployment — how it is introduced and governed — rather than to anything intrinsic to the software. That is what the research supports. Bring it in deliberately, as a source of instant feedback, individual support and clearer explanations of hard material, and the educational advantages are substantial and well evidenced. Bring in the same tools with no guardrails, no structure and no training in AI literacy, and you get dependency, blunted critical thinking, and a blurred line around whose work an assignment really is. A 2026 Brookings Institution report, widely cited since, found that under prevailing usage patterns the risks currently exceed the benefits — while taking care to note this is not a foregone conclusion, and can shift with sounder policy, training and safeguards.

02Why the Answer Refuses to Be Binary

Nothing called "education" happens uniformly across the world. Picture a five-year-old with an AI reading tutor, an undergraduate having ChatGPT draft an essay, and an instructor using AI to mark quizzes faster: three unrelated scenarios, different risks attached, different levels of maturity in play, and no shared notion of what a good outcome would even be. Study just one of them and you arrive at a different conclusion than you would from studying another.

Add to that the age of the evidence itself. Widespread student access to generative tools dates back only a few years, and nobody has yet finished the decade-long work needed to say how continuous use through the formative years shapes skills across that span. Neither the worries nor the gains should be waved away on those grounds — but the present picture does call for humility instead of confident pronouncements either way.

03Where the Benefits Are Strongest

None of this is speculative. Independent studies and student surveys keep turning up the same pattern, and the advantage learners name most often is precisely this practical kind of help — walking through exercises, building revision aids, and unpacking ideas that had not clicked.

04Where the Damage Shows Up

The misgivings are every bit as grounded. Over a year, the Brookings Institution's Center for Universal Education ran a global study spanning more than 50 countries; its 2026 report concluded that under present usage patterns the dangers generative AI poses to learners outweigh what it delivers. Named among the findings were threats to student privacy and safety, erosion of trust in learning as a process, deepening technological dependence, and the prospect of widening the gap between students whose homes offer differing levels of support and oversight.

Ask students directly and comparable anxieties surface: integrity worries, doubt about whether the information a model hands back is accurate, a sense that their own problem-solving muscle is atrophying, plus legitimate questions about data privacy and bias. One fear recurs in particular — that genuine work of theirs will be misidentified as machine-generated, a very concrete worry sitting underneath the grander argument.

05Reading the Evidence as a Whole

Bring the two sides together and a rough scholarly consensus emerges. AI can genuinely lift efficiency, accessibility and personalisation when it is designed with a purpose and anchored in sound pedagogy — while unsupervised use poses real hazards to critical thinking, academic honesty and fair outcomes. Neither finding comes from a single lab; they recur across research groups, which is itself informative. This is not one camp inventing a benefit or a danger out of nothing.

It is also worth noting that researchers and students do not weight these findings identically. Comparative work finds learners generally more willing and more ready to adopt the tools, while instructors more often foreground the ethical hazards and the possible damage to how well people learn. Neither group is simply mistaken; they tend to be examining different portions of one picture, from positions that carry different duties.

06Deployment Practice Outweighs the Software Itself

Nearly every study in this area converges on one point: results turn on how AI is brought in and structured, not on any fixed quality of the technology. The Brookings report organises this around three pillars — students prospering through quality teaching, education systems preparing via genuine AI literacy and professional development, and students being protected by concrete safeguards covering privacy, safety and emotional wellbeing.

This follows a pattern that shows up whenever a potent new tool reaches the classroom: the instrument is seldom what decides the outcome — the scaffolding around it is. The same logic runs through the argument over whether AI skills belong on the school curriculum at all; the answer tends to depend far less on the technology than on how deliberately it is taught and overseen.

07What Students, Parents and Teachers Can Do

1. 1

   Let AI test your grasp, not do the work

   1 Attempt the problem yourself first, then bring in AI to check or unpack your reasoning; the learning survives and the help still arrives.

2. 2

   Say openly how AI was used

   2 Saying plainly where AI figured into a piece of work, rather than concealing it, sidesteps integrity trouble and sets an honest example.

3. 3

   Write rules that are specific

   3 A policy that is fuzzy, or missing altogether, generates more confusion and risk than guidance spelling out, subject by subject, what is allowed.

4. 4

   Pair AI access with AI literacy

   4 Explaining how the technology functions, where it breaks down and how to judge what it produces counts for as much as handing out the logins.

5. 5

   Notice the warning signs of dependence

   5 A learner who can no longer do comparable work unaided has sent a signal: pull the tool back and rebuild the underlying skill head-on.

08Ways People Get This Argument Wrong

• Framing it as one question with one answer. Under the label "AI in education" sit tools, ages and scenarios so varied that they call for separate verdicts.
• Waving the dangers away as exaggerated. Research genuinely supports the worries about critical thinking and integrity; they are not merely institutional nerves.
• Writing off the upside as marketing. Personalized feedback and accessibility improvements are documented just as thoroughly — they are not merely sales copy from ed-tech vendors.
• Overlooking how much age and setting matter. What suits a graduate student does not follow automatically for a ten-year-old.
• Holding off until "the research is settled". Classrooms are already adopting this technology at speed, which makes considered guardrails today worth more than perfect certainty somewhere down the line.

Curious about the larger claim — just how consequential this technology is outside school walls? Our article on just how big a deal AI is beside the internet weighs that proposition in the same even-handed way. And if unease about safety and control is part of what makes you hesitate over AI in schools, our look at whether open models are a risk follows a neighbouring strand of the same discussion.

09Questions Readers Ask Most

10The Bottom Line

Back to the opening question. On the evidence available, the fairest verdict is that AI is both — and which tendency dominates in a particular classroom has far less to do with the technology than with how it is deployed, supervised and understood. The gains are concrete: speedier feedback, support tailored to the individual, and meaningful access improvements for those who need them. So are the costs: dependency, eroded critical thinking, and integrity problems that many current systems are not built to handle well.

Waiting for a clean verdict to arrive is the less useful option. For learners, parents and teachers alike, what actually shifts the odds toward benefit is a set of habits: being open about how these tools are used, writing rules that mean something, building real AI literacy, and checking regularly whether the tool is deepening your understanding or standing in for it. Ask that consistently and it answers the question better than any single study could.