Analysis of the ability of artificial intelligence tools to generate student-style code and detect such generated code

Abstract

It is known that artificial intelligence tools (AIT) can generate program code in different styles: the general style, the styles of other AITs, the teacher's style, the style of a professional developer, etc. But if a first-year student provides code for a task that was generated using an AIT in some such professional style then the teacher can easily determine the non-independency of performing such work. The article studies the possibility of AITs to generate code in the "first-year student" style and the ability of AITs to determine that this code was generated using an AIT, and was not actually written by a first-year student. The aim of the article is to investigate how easily a first-year student can mask the use of AITs when writing programs in C language for educational tasks in the disciplines of the first-year study. This study was conducted for the code generated in the student's style by the latest free versions of three AITs, which are currently among the most popular and most used by students: ChatGPT, Copilot, and Gemini. As a result of the study, it was found that the AITs ChatGPT, Copilot can mask their code as the student's code quite well, while the AIT ​​Gemini did not cope with this task. In any case, this applies to programs written in C language for tasks of initial complexity of the first year of study. The conclusions summarize the results of the study and make recommendations to teachers regarding the use of the studied AITs. The following are noted as areas for further research. Since the power of AI is developing explosively quickly, it seems appropriate to repeat the research of this article again in a year or two. Another direction can be proposed to conduct a similar study using other programming languages, as well as when performing tasks in other disciplines and/or of greater complexity