Explainable Artificial Intelligence in Automated Essay Scoring: A Systematic Review of Transparency In AI-Based Educational Assessment

Abstract

Background: Artificial intelligence (AI) is increasingly being used in educational assessment, particularly through automated essay scoring (AES) systems that apply machine learning and natural language processing to evaluate textual features such as grammar, coherence, vocabulary, and semantic relevance. While these systems offer efficiency, scalability, and reduced grading time, they raise significant concerns regarding transparency, interpretability, and fairness due to their “black-box” nature. This lack of explainability can undermine trust and accountability in educational decision-making.

Methods: This study conducts a systematic review of the literature on explainable artificial intelligence (XAI) in AES. It examines existing AI models used for essay scoring, identifies commonly applied explainability techniques such as SHAP, LIME, and attention-based visualization, and analyzes their role in improving transparency and interpretability across interdisciplinary domains, including educational technology, natural language processing, and learning analytics.

Results: The review finds that although AES systems are effective in replicating human-like scoring, most models lack transparency in their decision-making processes. The integration of XAI techniques has shown promise in providing insights into how models assign scores based on textual features. However, research in explainable AES remains fragmented, with limited standardization and varying implementation approaches across studies.

Conclusion: Explainable AI has significant potential to enhance transparency, trust, and fairness in automated essay scoring systems. However, there is a need for more cohesive and standardized research efforts to fully realize its benefits. Future work should focus on developing unified frameworks and evaluating the broader educational impact of explainable AES systems.