PREVENTING THE MASSIVE USE OF NEURAL NETWORKS IN THE PRACTICE OF TEACHING HUMANITIES DISCIPLINES

Abstract

In the context of the rapid development of neural networks and mass access to them, the question arises not only about their productive use in higher education, but also about the possibility of preventing their use in situations, in which students must develop their competencies independently. For many typical tasks (write a text on a given topic, find answers to test questions, etc.), modern students use neural networks, thereby not developing the competencies necessary for obtaining an education. In this regard, the purpose of the study is the scientific substantiation and implementation of a practice-oriented task, which at the present stage of development of artificial intelligence cannot be performed by it. In connection with the widespread use of neural networks and artificial intelligence, the question of not only interaction with them in the educational process, but also the question of countering their malicious use is relevant. This study proposes a task algorithm that allows students to stop the uncontrolled use of artificial intelligence. The corresponding technology can be transferred to other disciplines of the humanitarian cycle. The study used a task on the theory of language, which was proposed to be solved by the Chat GPT 4 and Claude 2.1 neural networks. As part of this study, it was demonstrated that neural networks cannot cope with the correct solution of tasks related to the connection of descriptions and any linguistic units, e. g. words or letters. Know-how in this case is the formulation of tasks in the first person, as if on behalf of the specified language unit. Testing tasks to connect two parts of a table using two different neural networks show their helplessness in terms of correct answers, i. e. errors ranged from 100% of cases to 30%. The proposed technology can be extrapolated to other disciplines of the humanities.