Artificial Intelligence Models and Test Performance

(Case Study in Accounting Program)

Abstract

Integrating artificial intelligence (AI) in education has required considering the conservative approaches of teaching and assessment systems, as AI impacts how educators operate, how students learn, and how learning outcomes are evaluated. The primary goal of this research paper is to compare AI tools’ performance with students’ performance through experiments on the final examination tests in the accounting program. In the conducted experiment, ChatGBT achieved a performance rate of 77 percent in a series of tests that encompassed the final examination, which included 11 courses and a total of 100 questions. The performance of the top-scoring students exceeded the results of the Copilot and Deepseek models; however, it did not surpass that of ChatGBT. The study findings have shown that ChatGPT performed less effectively on multiple-choice questions that require extensive calculations and accounting estimations; however, it was good at simpler multiple-choice and matching tasks. Thus, it was concluded that permitting students to undertake online examinations in unsupervised settings or utilize AI tools without limitations can enhance student performance in unreal ways while simultaneously minimizing reliability in assessment outcomes.