“Help in the hands of a professional”: philologist on the use of generative neural networks in schools and universities

A number of countries are discussing the need to introduce mandatory labeling of content created by artificial intelligence. At the same time, many schoolchildren and students use neural networks to complete educational tasks, which causes concern among both parents and teachers. And designers, writers and journalists fear that they will not be able to compete with generative technologies in the future. Candidate of Philological Sciences, Associate Professor of the Department of English Philology at the Institute of Foreign Languages ​​of Moscow State Pedagogical University Marina Ovsyannikova explained in an interview with RT that these fears are groundless, because only a person can create truly significant works. At the same time, neural networks can become a very useful aid in studying, the specialist believes.

— China is discussing the introduction of mandatory labeling of content generated by AI, since this technology allegedly creates new challenges to public and national security. When it comes to working with texts, is there any real cause for concern?

— As a rule, people are afraid of new technologies, this is normal. For technology to work for good, it must be used correctly. And this is exactly what needs to be taught to schoolchildren and students; this is the task of the education system. I don’t think that AI currently poses any threat and its “works” need to be specially labeled. Although it is possible that this will be required in the future.

— How exactly do advanced generative neural networks create texts? Is this similar to the human writing process?

— Any neural network is based on a mathematical model, the organization of which resembles biological neural networks that are present in the human brain. However, we should not forget that the brain functions much more complexly than any modern neural network.

Generative neural networks, which can generate texts, represent a large probabilistic model of language. It is trained on large data sets to predict the next word in a sentence, as well as the sentence in a text. This is similar to the old T9 program in phones. Modern neural networks can generate new content based on ready-made samples on which they were trained. At the same time, of course, the principles of human text creation are much more complex.

Gettyimages.ru Patrick Pleul

“Today, students can easily not only “copy” papers on the Internet, but generate them using neural networks. What signs can be used to determine that a text was written by AI?

— A text written by a person is always more complex syntactically. For example, depending on the style of the text, we can use repetitions, elliptical constructions when some member of the sentence is omitted. Sometimes we use special punctuation that implies our emotions. Neural networks can’t do any of this yet. In addition, from a lexical point of view, the neural network has rather low vocabulary variability. And at the same time, some rarely used word may suddenly appear in the generated text, which also betrays its artificiality.

In terms of meaning, one should immediately be wary of an overly logical narrative. This is another feature of the generated texts. In addition, in such texts there are no evaluative implicatures – that is, when information is conveyed not directly, but hidden, in an indirect form, as a hint. This presentation is characteristic of human thinking, but is inaccessible to neural networks, just like sarcasm.

— What other practical advice can be given to teachers so that they can immediately understand that this is a digital product?

— An experienced specialist who constantly works with texts usually has linguistic intuition. First of all, the excessive “correctness” of the text can be alarming: it resembles a wall through which it is difficult to break through to the meaning. Banal comparisons, clichéd phrases, monotonous syntax. Teachers needed such skills even before the advent of neural networks, because schoolchildren and students always cheated. Previously, we were similarly alarmed by the appearance of atypical vocabulary and phrases in students’ texts. After all, each person has his own style, and if we work with him for a long time, we can immediately guess the author of an essay or other work.

In general, artificial intelligence can only imitate human thought. As the Austrian philosopher Ludwig Wittgenstein said, “The limits of my language mean the limits of my world.” A neural network does not have its own “world”; it is simply a mathematical model. You and I think, write text here and now, in the conditions of true reality, feeling it with all five senses, experiencing emotions.

By the way, another sign of a “digital” text is the lack of emotional connotation or, on the contrary, excessive emotionality. In general, it feels unnatural.

Gettyimages.com Johner Images

— In general, is the use of AI in education acceptable? If yes, in what cases and under what conditions?

— Moscow City Pedagogical University was the first to legalize the use of artificial intelligence technologies for students in the preparation of final qualifying papers. But here it is important to understand that AI is only a tool that can make the work easier, but not do it for the student. We teach how to use it correctly in our master’s program “Text Technologies: Creating and Editing Texts in Foreign Languages.” For example, we have a course on Neural Networks in Linguistic Research, where we teach students how to use AI tools to analyze language and literature in combination with traditional methods.

A neural network is a good help in the hands of a professional; it expands his capabilities. So, for example, using a neural network you can quickly select all precedent names (well-known names – RT) in the text. This will significantly speed up the work compared to traditional continuous sampling methods, when you sit, read the text and write down the necessary words. The neural network will do this very quickly and will also explain the cultural background and historical subtext of each name. But, of course, you need to remember that all AI-generated data needs to be verified. This is one of the main conditions for the correct use of neural networks.

Generative neural networks can also be used as a brainstorming tool. Artificial intelligence can sketch out a list of some ideas, even if they are usually banal, but they can give you a worthwhile thought. So, using the process of elimination, you can come to some interesting hypothesis.

— At what age are neural networks relevant and useful in the educational process?

— Working with generative neural networks structures our thinking very well. Therefore, in order to get a good return from AI, you must clearly understand in advance what kind of result you need, for what specific purpose you are using the neural network, otherwise there will be no result, it will not be possible to work intuitively and situationally with this technology. And of course, you need to have sufficient data and professionalism to evaluate the results obtained. And this will determine what industrial statement (request – RT ) you make for the AI.

Therefore, in my opinion, it is too early for younger schoolchildren who are still learning to structure their thoughts to resort to the help of neural networks. In higher grades, you can begin to learn how to use AI and develop these skills. And when used correctly, it will also help develop critical thinking skills, when a person will have to first evaluate the results himself.

— It turns out that if a person does not have the ability to think structurally and critically, then he will not be able to properly use neural networks?

– Right. I think that the widespread introduction of neural networks can even partially contribute to the development of critical thinking in people from an earlier age. Because working with them correctly requires constantly questioning and checking many things.

Gettyimages.ru

— If children get used to such content from an early age, won’t it turn out that the criteria for evaluating literature, for example, will shift and no one will be able to distinguish digital novels from the work of a talented writer?

— I understand these concerns, but let me remind you that great literature has already survived many historical and technological changes. I don’t think that digital texts will ever be perceived as examples of literary excellence. In general, in order to compile low-quality reading material, it is not necessary to be a mathematical model or a neural network. People themselves cope well with this simple task – there have been many such products at all times. In this we are equal with neural networks. But only a person is capable of creating a truly great work. Talent will remain talent, and the creator will remain a creator.