IDENTIFYING NOVICE PROGRAMMING MISCONCEPTIONS AND AI UNDERSTANDING GAPS AMONG KAZAKHSTANI 7TH-GRADERS

Abstract

This article presents a qualitative and quantitative analysis of typical syntactic, logical, and conceptual errors, as well as naive mental models of artificial intelligence among 7th-grade students (N = 24) who completed a 34-hour elective course "Python Artificial Intelligence Programming." Unlike summative evaluations of course effectiveness, this study focuses on the internal mechanisms of student difficulties during the transition from basic Python syntax to applied libraries (NumPy, Pandas, OpenCV, scikit-learn). Based on the analysis of 120+ saved digital code artifacts (Thonny IDE log files and GitHub Classroom), tests, and semi-structured interviews (n = 8), a systematic error classification was derived. It was established that the most frequent problems were array dimension mismatches (70.8%), confusion between return and output statements (return vs print, 54.2%), and input data types (37.5%). Regarding AI perception prior to the course, 79.2% of students demonstrated anthropomorphic misconceptions ("AI thinks like a human"), which shifted after the course to an understanding of AI as a mathematical model trained on datasets (in 91.7% of students). The findings form a foundation for adapting instructional materials and reducing cognitive load in K-12 IT education.