Technology Popular

83% of students who used AI could not quote a sentence they had just written

Three groups, one essay task, electrodes on their heads. The memory test afterwards is the part worth reading twice.

83% of students who used AI could not quote a sentence they had just written

Using AI tools for creative work such as writing affects how people think and learn, according to researchers at the MIT Media Lab in the United States.

Over several months they studied 54 students and found that, compared with people who wrote essays unaided, those who relied on AI tools showed weaker neural connectivity in the brain and weaker memory.

How the study was built

Students from the Massachusetts Institute of Technology, Wellesley College, Harvard University, Tufts University and Northeastern University took part.

They were divided into three groups and asked to write essays on set topics. One group used ChatGPT, the second used a conventional search engine, and the third used no technology at all.

In a fourth session the design was flipped. Those who had previously used AI were asked to write without any tool, while those who had started unaided were allowed to use AI.

Unaided, search, AI
Strongest to weakest brain connectivity, in that order. Photo: OIST from Onna Village, Japan, Wikimedia Commons.

What they measured, and how

While the students wrote, researchers used electroencephalography (EEG) to measure brain activity. They then analysed the essays using natural language processing, and interviewed the participants about their own experience.

The pattern of brain connectivity differed according to which tool was used.

The strongest neural engagement was found among participants writing with no external help. Search engine users showed moderate engagement. The students who relied on AI tools while working showed the lowest neural connectivity of the three.

The relationship was consistent: the more external help a participant took, the less their brain activity. That is a dose effect, not a simple yes-or-no finding, and it is the more interesting result.

The memory test

The researchers also checked how well participants could remember what they had written.

Immediately after finishing, students were asked to quote a few sentences from their own essay.

83.3% of the participants who had used AI in a session failed to quote their essay correctly. The search engine group and the unaided group remembered far better.

Read that again with the timing in mind. Not a week later — minutes after writing it.

What this does and does not show

Fifty-four students at five universities is a small sample, and the study measures one task over a few months rather than long-term learning. It does not show that AI makes anyone permanently worse at thinking.

What it does suggest is narrower and more useful: text that a tool produced on your behalf does not get stored the way text you assembled yourself does. If the goal is a finished document, that may not matter. If the goal is to learn the material — which is the whole point of a student essay — then the finished document was never the point.

The practical reading for anyone studying here: use the tool to check and to question, and write the sentences yourself. The difference shows up in what you can still recall an hour later.

There is a cheap way to test this on yourself, and it needs no equipment. Finish a piece of work, close it, and try to say what you argued without looking. If you cannot, the document exists but the learning did not happen — and in an exam hall, only one of those two is available to you.

Source: Prothom Alo

Written by

Tech BD

Tech BD is an independent technology publication based in Dhaka, published in English and Bangla. Articles appear under the masthead rather than under individual bylines; corrections and source questions reach the editor at info@tech-bd.com.