Homogenization and Epistemic Dependency: How Generative AI Threatens Task-Based Language Assessment in ESP Contexts

Abstract

Task-based language assessment assumes that the learner’s performance reflects the learner’s own developing communicative competence. Generative AI challenges this assumption by allowing students to delegate task completion to an external system, thereby threatening task authenticity, performance authenticity, and the interpretability of form. The study aims to examine how 40 university English for Specific Purposes teachers at three Ukrainian universities experience the impact of generative artificial intelligence on the task-based language teaching and assessment validity chain. This issue is significant since, in the absence of institutional AI policy, teachers are the primary regulators of AI use in assessed contexts, and their decisions are the primary mechanism through which validity is maintained or compromised, yet remain empirically under-examined. The study employed a qualitative approach, including analytical content analysis of open-ended survey responses and structured classroom observations of 12 teachers (a purposive subsample). The results were five-fold: teachers found artificial intelligence useful for task material generation and formative feedback; teachers had consistent validity concerns for summative assessment; formative-summative asymmetry emerged in different disciplinary clusters; teachers independently indicated that students’ output converged towards a single artificial intelligence-modified output; and students would not be able to critically evaluate the artificial intelligence-generated text. Our findings are interpreted using the Task-Based Language Teaching-Assessment-Artificial Intelligence Nexus, a preliminary heuristic model that identifies three interconnected nodes of validity threat. The study argues that this suggests the need for formal AI policy, and for teaching AI-critical genre literacy in English for Specific Purposes education.