Can artificial intelligence help in learning: American researchers conducted an experiment
Students who used artificial intelligence had false confidence in their knowledge
Boris Jovanovic/Getty Images
Scientists assume that the use of artificial intelligence during education can become a “crutch” for schoolchildren, because in this way they do not get quality knowledge.
Researchers from the University of Pennsylvania (USA) conducted a corresponding experiment in which almost a thousand high school students from Turkey participated. writes Popular Science.
All schoolchildren attended school mathematics classes, but in the after-school class they independently completed the tasks.
The scientists divided the research participants into three groups – the first prepared independentlythe second group used ChatGPTand some of the schoolchildren studied using a different version of ChatGPTwho gave hints for solving problems.
*ChatGPT is a chatbot with artificial intelligence designed to provide answers to questions in a dialogue format.
Schoolchildren from all groups performed the same tasks on different topics during four cycles of classes, each of which lasted 90 minutes.
Subsequently, the researchers analyzed the correctness of the solutions obtained by the participants of the experiment.
It turned out that the group with access to AI performed correctly by 48% more practical tasks than schoolchildren who did it independently.
On the other hand, students who were prompted by a tutor on the basis of AI to solve problems had 127% more correct answers.
Subsequently, the authors of the study conducted additional testing to check how the students eventually mastered the knowledge of the studied topics. This time, schoolchildren solved the problem without the help of AI.
It turned out that the second group had 17% worse resultsthan those who studied independently.
In addition, students who were not assisted by artificial intelligence outperformed the group that was given tips by a “tutor” based on the ChatGPT chat.
Researchers suggest that even artificial intelligence imitation of a teacher can “significantly interfere with learning.”
The participants of the experiment often did not try to understand the topic, but simply asked for an answer, so the authors of the work believe that schoolchildren use the chatbot as a “crutch”.
As a result, they probably did not develop problem-solving skills like those who did it on their own.
According to scientists, another factor of negative impact on the overall results of students who used AI was that the bot made incorrect arithmetic calculations in 8% of cases.
A step-by-step approach to problem solving was false in 42% of cases. Instead, errors in the “tutoring” version of artificial intelligence were minimized.
The researchers also conducted a survey of the experiment participants. It turned out that the use of AI has given rise to false confidence – schoolchildren believed that they coped with the test much better than it actually was.
Scientists published a preliminary version of the research paper on SSRN in July 2024. However, the article has not yet undergone the necessary peer review before publication in the journal.
Previously, researchers from the British University of Reading sent AI-generated exam answers to your own professors. The teachers did not suspect fakes and rated the works higher than the students’.
