Using AI to do homework and assignments reduces learning by 20%

Using AI to do homework and assignments reduces learning by 20%

Technology offers students a temptation that is hard to resist: getting better grades with less effort. But that shortcut can take a toll on learning.

Read more The Pentagon withdraws the aircraft carrier from the Persian Gulf where a deterioration in living conditions was reported

The use of ChatGPT and other brands is already widespread among teenagers and young people who use it to replace their cognitive effort or also to enhance their learning. This practice is catching schools (and universities) off guard. Educational administrations around the world are exploring ways to adapt, ranging from a return to oral exams to the integration of AI in the classroom.

Meanwhile, educational practice has barely changed. Teachers are not sure that the exercises they assign help improve or consolidate what is learned in class if the one answering is a machine. They do not know if the student is on the right track or needs help. And they fear that the evaluation grade may not be fair. How to assess high school research work if there is suspicion that it was written by ChatGPT?

The grade of homework for students who used AI was 18% higher than that of their classmates

Above all, the worst part is that cheating harms the student themselves because they may pass, even with a good grade, but they will not have learned that knowledge.

Until now, the impact of AI use on long-term learning had been little quantified. A study, summarized this week by the weekly The Economist, provides concrete figures. The grades of students who use AI improve by 18% compared to their desk mates who do not use AI. Those same students perform 20% worse than the rest in exams.

Researchers David Stromberg, from Stockholm University, and Victor Lei and Wu Yanhui, from the University of Hong Kong, followed 27,000 students aged 12 to 18 in China for 30 months between 2022 and 2025. About 80% of the students in the sample reported using models like Doubao and DeepSeek.

The researchers collected their grades, the time they spent studying nine subjects, as well as the results in monthly exams and in external evaluations to the school, such as entrance exams to high school or university.

Estonia is developing artificial intelligence for its educational system

After six months, the average grade of homework for students who used AI had increased by 18% across all subjects. Their study time at home had decreased from 64 to 45 minutes on average.

However, in exams where they do not have the support of technology, they obtained worse results than their classmates. It was observed that the failures were not immediate but accumulated over time. The gap increased after six months of using AI in classroom tests. Also in end-of-secondary-stage evaluations (to access high school) or university entrance exams (equivalent to the Spanish selectivity). In these cases, the decline is 18% and 24%, respectively.

Read more A magnitude 7.2 earthquake shakes the Andean region of Ayacucho, in southern Peru

Learning losses are greater in social science subjects (27%), followed by STEM disciplines (science, technology, engineering, and mathematics) with 22%. In languages, the setback in English is 17% and in Chinese, 9%.

The research identifies the most affected groups. They are younger students, those in the stage equivalent to Spanish ESO, as well as high-performing students. The effect is also greater among boys.

Despite these data, this trend lessens over time as if students had learned to use it better, according to the researchers. After two years, the loss was no longer 20% but 16%.

The pressure on Chinese students

The research sample is based on Chinese students who, as the researchers recall, are under strong social pressure regarding their academic performance. Schoolchildren in this country dedicate 55 hours a week to studying, combining classroom and home, well ahead of the OECD average of 44. Class sizes range from 50 to 60 students. Over the six years of secondary education, they study three core subjects (Chinese, mathematics, and English) and several complementary subjects (physics, chemistry, biology, history, politics, and geography). They prepare for two decisive exams: the entrance exam to secondary school (zhongkao) and the university entrance exam, the dreaded gaokao, in which almost 13 million students participate.

When analyzing the sample details, Stromberg, Lei, and Yanhui saw that not all students used AI in the same way. 20% of students had good exam results. It coincided that those students spent an hour doing homework, the same time as their classmates who did not use chatbots. The researchers deduce that they used the technology not to get the homework done quickly but as a personal tutor to resolve doubts.

These results reinforce a widespread concern in educational systems worldwide: how to prevent AI from replacing the effort a student needs to learn. In early August, Denmark issued measures to curb cheating that will be applied in the school year starting now. “Unfortunately, we have a problem with students using AI to cheat,” said Education Minister Magnus Heunicke. “It is necessary to act now.” Thus, students will be required to orally defend their work and tasks will be done at school, not at home, and under supervision.

The response of some countries is a return to oral exams or homework at the educational center

UNESCO tries to overcome the debate between banning or accepting. It proposes that AI become an educational competency. In general, in most countries, such as Spain, administrations are developing specific national frameworks on when, how, and for what AI is used. France has established usage rules according to students’ ages and has introduced mandatory AI training.

China and Estonia are countries at the forefront. The Asian country approved a specific plan last April to integrate AI at all educational levels by 2030. Estonia, for its part, works with tech companies like OpenAI and Google to develop educational AI adapted to its system, capable, for example, of using the Socratic method. It does not offer an immediate answer but guides the student through questions so that they reach their own conclusions. It aims to ensure that AI is used so that students continue learning.

Read more Carlos Alcaraz announces his return to the courts at the US Open: “I’m back”

Translated from

Leave a Reply

Your email address will not be published. Required fields are marked *