雅思阅读 15: Homework Up, Exams Down（作业涨了，考试跌了）

改编自 The Economist（2026年8月）。雅思阅读 Section 3 难度，约 1050 词。 素材来源：The Economist / 27,000-student China study by Stockholm University & HKU

Reading Passage

A. When The Economist reported in August 2026 on a study tracking 27,000 Chinese secondary-school students, the numbers it published looked at first like good news for educational technology. Students who used generative AI tools such as Doubao and DeepSeek for homework saw their average coursework scores rise by 18 percent over a six-month period, while the time they spent on assignments fell from 64 minutes to 45 minutes per subject. The headline that spread across social media read: "AI makes homework faster and grades better." But the second number told a different and far more troubling story. When the same students sat invigilated, closed-book examinations — with no access to their AI tools — they scored 20 percent lower than classmates who had not relied on artificial intelligence. The juxtaposition was sharp enough to unsettle both educators and parents. Here was a technology that, by every conventional measure of classroom output, seemed to be doing its job — yet the students exposed to it were demonstrably less able to reason through a problem on their own.

B. The research, led by David Stromberg of Stockholm University with Victor Li and Wu Yanhui of the University of Hong Kong, followed students aged 12 to 18 in an environment marked by rapid AI adoption. Roughly 80 percent of participants reported using AI models during the study period; the remaining 20 percent formed a control group. Over six months, homework scores for AI users climbed by 18 percent across all subjects. But invigilated tests produced the inverse result: AI-reliant students underperformed their non-AI peers by a full 20 percent. The pattern held across subjects and school types, and it was strongest among students who used AI most frequently. Perhaps most strikingly, the historical correlation between homework grades and examination performance — a relationship that held for decades — had reversed: students with the highest homework scores were now, on average, the ones who performed worst on closed-book tests. Because the sample was drawn from schools across urban and rural regions, and because it tracked the same pupils over an extended period rather than comparing unrelated groups, the researchers considered the result unlikely to be a statistical accident. Even so, correlation across a single cohort, however large, leaves open the possibility that other traits — motivation, family income, prior attainment — partly explain who reached for the tools.

C. The findings align with a 2024 experiment at the University of Pennsylvania, in which mathematics students practised using notebooks and textbooks while comparison groups used ChatGPT and intelligent tutoring software. The technology users outperformed their peers in short-term practice sessions but lost that advantage on subsequent closed-book examinations. The mechanism, researchers suspect, is not that AI "hampers" learning directly, but that it short-circuits the productive struggle that underpins durable understanding. When a student can generate a complete solution within seconds, they never experience the confusion, error and revision that build the cognitive structures needed to reproduce that solution unaided. Coursework scores rise because the output is correct; examination scores fall because the underlying learning never happened. As the Brookings Institution cautioned in a separate guidance paper, excessive reliance on technology as a substitute for thinking limits the development of the cognitive skills that closed-book assessments are designed to measure. This is not a new anxiety. Decades of research on worked examples and cognitive load have shown that learners retain more when they are made to retrieve information from memory than when they are simply shown it. AI, in this light, is an unusually capable version of a well-known danger: the easier the answer becomes to read, the less learning the reader actually does.

D. The study's implications extend beyond the classroom. Surveys in wealthy nations already show that 80 percent of university students use AI in their studies, with figures reaching 94 percent in Britain and 93 percent in Germany. Teachers report a sharp increase in formulaic, uniform essays suspected of being bot-generated. Until the Chinese study, however, rigorous evidence on the actual learning impact of these tools was scarce; most existing research was either small-scale, short-term, or funded by ed-tech companies with a stake in positive results. The Economist's report prompted intense debate on academic social networks, with commentators divided between those who saw the findings as a necessary warning and those who argued that the study had not yet undergone peer review and that its correlation did not prove causation. The research paper was indeed released before formal peer review at the time of publication. Some critics also noted that self-reported AI use, rather than logged software data, could misclassify heavy users as occasional ones, while defenders replied that the effect sizes were too large to vanish under such noise. Whatever its final reception, the study arrived at a moment when schools were already scrambling to draft rules they had little evidence to base them on.

E. The practical recommendation emerging from the data is not to ban AI but to change how it is used. The Economist quoted the authors as concluding that "AI can boost learning productivity, but only for those who use the technology intelligently." Students who were asked to attempt a problem independently before consulting an AI assistant showed smaller examination declines than those who reached for the tool first. The pedagogical principle is old: effort precedes insight. AI appears to be a tool whose benefits depend entirely on the cognitive state of the person using it. Used after independent struggle, it can serve as a tutor; used as a shortcut, it becomes a crutch. Whether schools will be able to enforce that distinction at a time when subscription costs roughly ten dollars a month and access is nearly universal remains an open question. Nor is the dilemma confined to teenagers. Universities, employers and professional bodies face the same question at every level of training: how to certify competence when the raw material of competence — a written solution, a solved equation, a reasoned argument — can now be outsourced. The authors do not pretend that a single study settles this. But they are clear that the easy path of letting AI do the work, however pleasing the homework returns it produces, carries a bill that comes due later, in the examination hall and beyond.

Questions 1-4

Choose the correct heading for paragraphs B, C, D and E from the list of headings below.

List of Headings i. What the data showed across 27,000 students ii. Why homework grades and exam scores diverge iii. The role of AI in university admissions iv. The wider context: surveys, essays and methodological debates v. How to use AI productively without losing understanding vi. A history of educational technology in China vii. Why peer review is always unreliable

1. Paragraph B: ____
2. Paragraph C: ____
3. Paragraph D: ____
4. Paragraph E: ____

Questions 5-8

Choose the correct letter, A, B, C or D.

5. What was the most striking finding? A. AI users spent more time on homework than non-users. B. The historical link between homework grades and exam performance reversed. C. AI tools were used by only 20 percent of students. D. AI improved exam scores by 20 percent.

6. Why did AI users perform worse on exams? A. They were more nervous during invigilated tests. B. They had skipped the struggle needed to build lasting understanding. C. The exams were more difficult than homework. D. They were not allowed to use calculators.

7. What limitation of the study is noted? A. It was conducted only in private schools. B. The paper had not yet undergone formal peer review. C. It lasted for two years. D. It excluded students who did not use AI.

8. What practical approach does the article recommend? A. Banning all AI tools in schools. B. Encouraging students to attempt problems independently before using AI. C. Replacing closed-book exams with open-book assessments. D. Making AI subscriptions mandatory.

Questions 9-13

Do the following statements agree with the claims of the writer?

Write:

• TRUE if the statement agrees with the information
• FALSE if the statement contradicts the information
• NOT GIVEN if there is no information on this

9. AI users' homework scores rose by 18 percent over six months.
10. AI users spent more time on homework than non-AI users.
11. The University of Pennsylvania experiment showed similar patterns.
12. The study was funded by an AI company.
13. The authors concluded that AI has no educational value.

Questions 14-15

Complete the summary below using NO MORE THAN TWO WORDS from the passage.

The study found that AI reliance raised (14) __________ scores but lowered closed-book (15) __________ performance, suggesting that AI is most beneficial when used after independent effort.

答案与解析

← 上一篇 | 返回雅思焦点 | 下一篇 →