雅思阅读 5: The AI Crutch（AI 拐杖）

改编自 André Barcaui / Social Sciences & Humanities Open（2026）。雅思阅读 Section 3 难度，约 1050 词。 素材来源：https://www.psypost.org/chatgpt-acts-as-a-cognitive-crutch-that-weakens-memory-new-research-suggests/

Reading Passage

A. When a student delegates a mental task to an external tool — using a calculator for arithmetic, a notebook for a grocery list, or a mapping app for directions — psychologists call it cognitive offloading. In moderation, offloading is harmless, perhaps even sensible. But generative artificial intelligence platforms such as ChatGPT have extended the concept in an unsettling new direction: they do not merely store information; they perform the thinking itself — analysing, synthesising, problem-solving and producing polished output within seconds. A randomised controlled trial conducted by Brazilian researcher André Barcaui now suggests that this convenience may carry a hidden cognitive cost. Students who used an AI chatbot to study new material retained significantly less of it six weeks later than those who studied the same material by traditional means — even though they finished their work nearly half as quickly. The study joins a small but growing literature warning that tools which think for us may, over time, diminish the mental operations they outsource. Earlier research on calculators and spell-checkers had already suggested that offloading routine mental work erodes the automatic fluency those operations once built; generative chatbots, critics now argue, extend that erosion from arithmetic and spelling to reasoning itself.

B. Barcaui, a professor at Fundação Getulio Vargas and the Federal University of Rio de Janeiro, recruited 120 undergraduate business students aged 18 to 24, none with formal training in computer science. The participants were randomly divided into two equal groups. Both received the same assignment: research basic artificial-intelligence concepts — covering ethics, societal impacts and technical foundations — over a period of up to two weeks, then deliver a ten-minute presentation. The first group was instructed to use ChatGPT freely as a study aid, asking it to explain concepts, generate examples and structure their talks. The second group was barred from using chatbots and relied instead on course notes, academic databases and conventional internet search engines. After delivering their presentations, the students resumed their normal academic lives for forty-five days, allowing natural memory decay to run its course. No feedback was given on the presentations themselves, so neither group knew how it had performed; this was deliberate, since post-hoc praise or criticism could itself have become a learning event. The forty-five-day gap was chosen to outlast the short-term rehearsal that students habitually deploy after an assessment, ensuring that whatever remained had been consolidated rather than merely retrieved from working memory.

C. The retention test that followed was deliberately surprising: a twenty-question multiple-choice examination designed to measure conceptual understanding rather than rote memorisation. Eighty-five of the original students returned. The results were striking. The traditional-study group answered 68.5 per cent of questions correctly, while the chatbot-assisted group managed only 57.5 per cent — an eleven-percentage-point gap that held even when the data were adjusted for the fact that the AI group had spent far less time preparing (3.2 hours on average, compared with 5.8 hours for the traditional group). The deficit was most pronounced for highly technical topics; the gap narrowed somewhat for less technical material such as ethics and social impact. Prior familiarity with chatbots made no difference: frequent users performed just as poorly as novices. The gap was also largest when questions demanded that students integrate ideas from several lectures at once, rather than recall a single fact in isolation. This pattern — poorer performance on integrative questions — is precisely what the desirable-difficulties framework predicts: students who had struggled to connect ideas for themselves had built the flexible memory traces that such questions reward, while prompt-and-response study had not.

D. Barcaui interpreted the findings through the lens of "desirable difficulties" — a well-established psychological principle holding that learning improves when the learner confronts a certain amount of productive struggle. Actively trying to recall a fact or puzzling through a difficult concept forces the brain to build stronger memory pathways. By delivering immediate, fluent answers, a chatbot may strip away precisely that friction. Traditional students, spending their extra time re-reading, self-testing and manually connecting ideas, were exercising the brain's retrieval networks; chatbot users, by contrast, were writing prompts and reading generated text — an activity that feels productive but may fail to consolidate knowledge. The resulting "illusion of competence", Barcaui warned, is one of the most insidious aspects of AI-assisted study: the software makes students feel they understand material they have never truly grappled with. The mechanism is not mysterious. Retrieval practice — actively pulling information from memory rather than passively reading it — has been shown, in dozens of earlier experiments, to strengthen memory more reliably than re-reading. A chatbot session, however fluent it feels, rarely requires the learner to retrieve anything; the learner's own recall is never tested until the exam itself, by which point the gap between familiarity and understanding has become invisible.

E. The study is not without limitations. Nearly thirty per cent of participants did not return for the delayed test, though dropout rates were comparable across groups. The sample was drawn from a single Brazilian business school, raising questions about generalisability to other disciplines, age groups and educational systems. Barcaui himself is an enthusiast for AI's responsible use and does not argue for abandoning the technology. His conclusion is more nuanced: AI should function as a co-pilot that extends human capability, not an autopilot that replaces effort. Students might attempt a problem on their own before consulting a chatbot for feedback, he suggests, preserving the struggle while still benefiting from the tool. The broader implication extends beyond education. If the convenience of outsourcing thought undermines the internal structures that thought builds, the question is not whether to use AI but how to use it in a way that leaves the learner cognitively stronger rather than weaker. Should the findings hold, they also carry consequences beyond the lecture hall. Knowledge workers who draft emails, reports and even software with large language models may, over years of use, encounter the same subtle erosion of fluency that the students displayed. The most practical response is probably not abstention but sequencing: attempt the work first, consult the machine second, and let the struggle precede the answer.

Questions 1-4

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

List of Headings i. Why immediate answers may undermine long-term learning ii. The design of a controlled comparison iii. Financial costs of AI tools in education iv. The retention gap revealed by a surprise test v. How chatbots have improved academic performance vi. Limits of the study and a balanced conclusion vii. The history of cognitive offloading

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 main finding of the delayed retention test? A. Students who used ChatGPT performed slightly better. B. Both groups achieved identical scores after adjustment. C. Traditional students outperformed chatbot users by a wide margin. D. The gap was largest for non-technical topics.

6. Why did Barcaui adjust the data for study time? A. To show that the chatbot group spent more time on technical topics. B. To demonstrate that the learning deficit was not caused by shorter preparation. C. To prove that ChatGPT users were more efficient learners. D. To confirm that traditional methods require less time overall.

7. The "illusion of competence" refers to the feeling that A. students are incapable of learning without AI assistance. B. AI-generated responses are always factually accurate. C. one understands material that one has never actively processed. D. traditional study methods are unnecessarily slow.

8. What does Barcaui recommend? A. Banning chatbots from all university courses. B. Using AI after attempting problems independently. C. Replacing chatbots with printed textbooks. D. Increasing the time students spend on presentations.

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. The chatbot group finished their assignments in less time than the traditional group.
10. Students who were already experienced chatbot users performed better than novices.
11. The retention test was announced at the start of the study.
12. Barcaui believes AI should be completely abandoned in education.
13. The study sample included students from multiple Brazilian universities.

Questions 14-15

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

The principle of (14) __________ __________ suggests that productive struggle strengthens memory. When a chatbot removes this friction, students may experience an (15) __________ of competence rather than genuine understanding.

答案与解析

← 上一篇 | 返回雅思焦点 | 下一篇 →