In total, more than 200 students and 20 teachers took part in the experiment. Half of the students, who were determined by blind selection, were asked to write a scientific and practical work on the chosen topic using neural network technologies, including ChatGPT, Ai search, Gerwin AI, Balaboba and TurboText services.

The teaching staff had to identify among the prepared works those in which artificial intelligence was used.

As a result, experts were able to accurately identify 96% of the works written by the neural network. For 4% of undetected generated abstracts, it was found that students spent a significant number of hours editing the result issued by the AI.

Among the criteria by which it was possible to determine the generated texts, 86% of teachers noted the stylistic and spelling features of the text - for example, exceptional literacy, the absence of typos, and a large number of self-repetitions. In more than 72% of cases, the work contained gross factual and logical errors. In more than 67% of the essays, the narrative deviated from the topic and contained "empty" expressions that did not make sense.

"In general, all the texts created by the neural network represented a compilation of theses and proposals from previously written scientific papers and works and did not contain original judgments and conclusions," the RTU MIREA specified.

The rector of the university, Stanislav Kudzh, noted that the results of the experiment indicate that the current level of development of neural networks does not call into question the quality of certification and verification of the level of knowledge of students of higher educational institutions.

"Despite the resonance of the topic, including in the media, artificial intelligence currently cannot replace the experience and qualifications of the teaching staff. Moreover, the neural network is unable to create a fundamentally new scientific or creative product, since its work is based solely on the processing of knowledge embedded in it by a person. The imperfection of the technology was also announced by the students participating in the experiment, Kuj noted.

Earlier, Shamil Magomedov, Head of the Department of Intelligent Information Security Systems at the Institute of Cybersecurity and Digital Technologies of RTU MIREA, told RT how to recognize photos of people generated by a neural network.