Comparative Literature Review: Generative Ai-Based Automated Feedback Versus Traditional Assessment In Vocational Programming Assignments

Abstract

The vocational education sector in informatics faces structural challenges in the form of disproportionate teacher-student ratios, resulting in programming assignment assessments that frequently rely on traditional unit testing mechanisms characterized by binary outcomes (running correctly or producing errors). This Systematic Literature Review (SLR) aims to compare the effectiveness of Generative Artificial Intelligence (GenAI)-based Automated Feedback systems with traditional assessment methods in vocational education. Using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, 19 primary studies were extracted from major academic databases. The synthesis results identified the phenomenon of “Procedural Convergence Without Pedagogical Equivalence,” in which large language models (LLMs) are capable of aligning scoring procedures with human experts at very high precision levels, with Intraclass Correlation Coefficients approaching 1.0, yet fail to replicate pedagogical sensitivity in diagnosing students’ fundamental misconceptions. Within vocational contexts, AI operates similarly to a Quality Control (QC) machine that detects syntactic anomalies with high precision, while human educators function as “Master Craftsmen” who are sensitive to students’ learning histories and instructional contexts. The findings position GenAI as an augmentative decision-support tool that effectively reduces educators’ administrative workload without substituting their critical role in providing pedagogical scaffolding.