British scientists: ChatGPT bot passes exams better than students

By Alimat Aliyeva

As British scientists have found out, artificial intelligence can take exams no worse than ordinary students, Azernews reports.

Using the ChatGPT bot, one of the leading applications using artificial intelligence, researchers from the University of Reading designed 33 "students" and sent them to take exams in psychology for those wishing to obtain a bachelor's degree at this university.

According to the results of the study, the results of the fake "students" were on average half a point higher than those of their real peers. And essays written by artificial intelligence are almost impossible to distinguish from those written by real people: 94 percent of them did not arouse suspicion from the inspectors.

The scientists shared the results of the experiment in an article published in the journal Plos One. In particular, it notes that they consider the figure of 6 percent to be overestimated: in reality, in their opinion, it will be possible to identify an even smaller percentage of works prepared with the help of AI.

During the study, exam answers and essays completed by AI for the modules of the first, second and third courses were submitted together with the works of ordinary students, while their authorship was not indicated.

Interestingly, the results of the fake "first and second year students" exceeded the results of the real ones, and according to the results of the exams in the third year, ordinary students showed higher results. According to the researchers, this is consistent with the understanding that artificial intelligence is currently having difficulty with more abstract reasoning. "To date, this is the largest and most reliable study of its kind," the BBC points out.