Polina Yachmennikova about Russian education in the web of neural networks

At the end of 2023, the updated national portal about the Russian language “Gramota.ru” for the first time identified the “word of the year” – and it immediately turned out to be the word “neural network”. “In an everyday, non-terminological sense, a neural network is perceived by speakers as a kind of living creature that can be taught, trained, and with which one can communicate,” the word was described on the portal. “Artificial intelligence” (AI), according to the observations of “Literacy,” “remains in the mass consciousness a complex, incomprehensible, unpredictable, and sometimes dangerous technology.” In the field of education, neural networks and AI have already mixed the cards. If in 2020, when lessons and lectures went remote due to the pandemic, everyone was afraid of the complete replacement of traditional formats with online classes, then in 2023 it became obvious that technology can completely replace teachers, sometimes at the same time as students.

In February 2023, RSUH graduate Alexander Zhadan admitted on social networks that his diploma was written using ChatGPT (a chatbot with artificial intelligence from OpenAI). According to Mr. Zhadan, the work took 23 hours and 112 chat messages, and the originality of the text after checking by Anti-Plagiarism (a Russian system for detecting text borrowings in scientific and qualifying works) was 82% (70% is enough for admission to defense). It was possible to defend himself, however, only with a C grade, but the situation caused a resonance in the media: the young man was invited to work and to thematic round tables, and at the same time it was proposed to deprive him of his diploma. This has not happened yet. At RSUH only stated, that plagiarism, which “decades ago was the main problem of universities,” has been replaced by neural networks and now it is necessary to learn to recognize their creativity. So far, the university has proposed “limiting access to the ChatGPT chatbot in educational organizations due to its possible negative impact on learning” – without, however, explaining how to do this technically. The head of the Ministry of Education and Science, Valery Falkov, said that the graduate “tested the system for strength,” and “universities need to be rebuilt.”

Experts also stated that the young man “created an important precedent” – more likely, however, with his sincere confession than with the fact of using a neural network for a diploma (obviously, many chose to prudently remain silent).

By the beginning of May, Antiplagiat reported that it had improved its verification algorithms, so now the system can recognize text generated by neural networks. But it is unclear whether such a text should be classified as illegal borrowing; there have been no instructions from above in this regard. And by the beginning of the academic year at Moscow State Pedagogical University, students were officially allowed to use chatbots to write their final papers, since “this will develop analysis skills, and the generated texts still need to be processed independently.” There has not yet been any unanimity on this issue among the university community.

Schools have it even harder. It seems like I’ve managed to get used to the numerous GDZ (so-called ready-made homework assignments) and collections of “100 essays for high school,” which have long since moved to an online format, but I just have to get used to neural networks. Access to “Anti-plagiarism” is not provided for teachers, and the Unified State Exam is also checked without it. Progress so far is average. Thus, at the all-Russian competition “Territory of a Modern Teacher,” as one of the tasks, teachers were asked to distinguish a student’s work from material produced by a neural network.

“More than half of the teachers” coped with the task—57%. This means that 43% did not recognize and would easily accept homework done by a chatbot.

However, the organizers thought this result was good. Moreover, they associated the ability to distinguish the text of a neural network from the work of a schoolchild with the fact that teachers have already figured out how AI generates text, which means they have mastered the new technology. Well, or, as sometimes happens on school tests, they were just lucky.

The neural networks were also lucky on the tests: the Yandex press service said in October that their generative neural network Yandex GPT 2 passed the Unified State Examination in literature with 55 points. At the same time, 40 is enough for admission to the university. It has not yet been possible to overtake real graduates: the average score in this subject is 64. So Yandex GPT is not yet applying for budget places in universities. “The development of neural networks does not pose a threat to the Unified State Examination system, since the use of gadgets during exams is prohibited by the Unified State Examination procedure and participants in real exams will not be able to use the help of artificial intelligence,” the press service of Rosobrnadzor reassured.

While the Russian State University for the Humanities proposed limiting access to neural networks, deputies, in the name of discipline, generally banned the use of “means of mobile radiotelephone communications” (that is, primarily mobile phones) during lessons. So that schoolchildren would definitely not be able to reach any networks. For the educational process, parliamentarians are confident that students will have enough school equipment, which schools, according to the Ministry of Education, actively provide – although not everyone gets it yet. Teachers and administrators, however, have concluded en masse that the law was “about ten years late” and all educational institutions have long been regulating the issue on their own. Most often, children and teachers manage to simply come to an agreement, and the phone remains quietly in the briefcase during class.

The education system would generally like to see AI and neural networks among its allies.

Thus, the head of Rosobrnadzor, Anzor Muzaev, promised in the fall that at the upcoming Unified State Examination, all work would be run through a neural network in order to eliminate the possibility of cheating and substitution. The department has been dreaming about this for several years, but so far they have used the system only in a test format – training and preparation for exams. And in the new strategic direction of the digital transformation of education, approved this year, designed to promote “digital maturity,” the need to involve “artificial intelligence expert systems to automate the checking of homework and planning educational programs” is generally declared in order to relieve teachers. True, it later turned out that they would be able to unload only “in the horizon of 2025–2030,” and even then only partially.

Russian President Vladimir Putin gives the green light to AI for students too. At a meeting with young scientists at Sirius, he said that the study of AI in domestic school education is “tough to handle”; all that remains is to build programs according to age. However, teachers are not entirely happy with the technological provision of schools, as they periodically stated in conversations with Kommersant – but perhaps neural networks will solve this problem.

Educators and officials like to mention that education is one of the most conservative areas that is difficult to modernize. Perhaps this is why educational reform never stops. But on the path of change, most often it is only clear that something needs to be changed – and the final goal of this process is sometimes difficult to discern. Even now, it is not clear whether educational officials will be able to employ AI in their service and protect children from the harmful influence of neuro- and other networks.

In any case, experience suggests that the younger generation is always a few clicks ahead and will “test the strength of the system” more than once with new ways to cheat.

Sometimes this will even certainly benefit science and advanced technologies. But the system, of course, will endure. And it will continue to change in different directions. And the way life is now is that there will be much less simulation in these changes than their designers and participants would like.