Assessment UT EFL Learner S

The growing use of artificial intelligence in automated writing assessment has intensified a central question in EFL writing assessment: whether AI-generated judgments can be regarded as ethically defensible and valid representations of writing ability when compared with human raters. This paper addresses that problem by examining the comparative role of human and AI raters and by clarifying the principal ethical and validity issues that arise when automated systems are used to assess multilingual writers. Methodologically, the study adopts a critical integrative review of the literature combined with logical-conceptual analysis. Rather than generating new empirical data, it synthesizes and interprets current scholarship through a three-stage analytical structure concerned with comparative performance, ethical risk, and validity limits. The findings indicate that AI can approximate human scoring under structured, criterion-explicit conditions, especially where assessment focuses on formalized linguistic features. However, this convergence weakens when evaluation depends on higher-order discourse judgment, contextual sensitivity, and interpretive flexibility. The analysis further shows that ethical risks cluster around fairness, opacity, accountability, agency, privacy, and academic integrity, while validity remains constrained by construct underrepresentation, construct-irrelevant variance, and the conditional nature of score interpretation across tasks, genres, and learner populations. The paper concludes that AI is most defensible as a supplementary, human-governed instrument rather than an autonomous replacement for human judgment in EFL writing assessment.

کلیدواژه ها:

automated writing assessment ، AI raters ، human raters ، EFL writing assessment ، validity and ethics