Application of Artificial Intelligence in Research Tasks in the General Physics Course

Abstract

The scientific justification is presented and the prospects for the application of artificial intelligence in solving research problems in the course of general physics are considered. The authors emphasize that artificial intelligence is used in many areas of modern production, education, and science. The development of modern machine learning technologies at the beginning of the 21st century provided a significant breakthrough in the implementation of artificial intelligence in the educational process. The process of introducing AI into the educational process of many countries has begun at the state level, even entire ministries have been created that deal only with this problem. Artificial intelligence in education can be applied in the form of interactive platforms and chatbots that provide students with instant answers to questions and can apply mathematical tools to significantly facilitate and accelerate the solution of physical problems. This is especially important for the   research tasks. But the question arises - how correctly AI copes with the tasks set. The work tested modern AIs such as ChatGPT-4turbo, DeepSeek, Copilot, Gemini 2.0, Claude, Grok regarding the reliability of solving research physical tasks. In order to systematize the results of the study, a classification of physics tasks was proposed, which were divided into groups according to the content and degree of generalization, the nature of the formulation, the method of solution and research, and the didactic purpose, depending on the requirements of the task. Three main types of tasks related to solving research problems are considered: tasks in which the research begins with the beginning of the solution; tasks in which an intermediate result obtained in the form of a functional dependence is investigated; tasks in which the final result of the solution is investigated. To do this, each of the considered AIs was offered 30 tasks for consideration, the list of which included tasks of each type from the above classification of research tasks and tasks of increased complexity. Based on the conducted research, it was found that the tested AIs, such as ChatGPT-4turbo, DeepSeek, Copilot, Gemini 2.0, Claude, correctly solve about 60 % of the research problems, and with an incorrect solution, the deviation from the correct answer ranges from several percent to several times. The Grok AI shows a slightly worse result both in terms of the percentage of correctly solved problems (50%) and in terms of deviation from the true result. Based on the nature of the interpretation of the answers, we can confidently say that it is AI with a developed reasoning ability (in our research, these are ChatGPT-4turbo, DeepSeek, Copilot, Gemini 2.0, Claude) that show the best results when solving research tasks and tasks of increased complexity. The results of the research showed the possibility of using AI to facilitate and accelerate the solution of research problems, provided that the results are verified, since today it can be considered that AI is still in the learning stage. At the same time, the role of the teacher is not decreasing, but, on the contrary, increasing.