When the Target Domain Changes: AI-Mediated Construct Drift in High-Stakes English Language Assessment

High-stakes English proficiency tests, such as TOEFL and IELTS, have historically relied on a core interpretive logic: performance under standardized, unaided conditions serves as a reliable proxy for academic communicative readiness. However, the rapid integration of generative AI into academic workflows—including reading, drafting, summarizing, and revising—threatens this extrapolation. The author defines this misalignment as 'AI-mediated construct drift,' where the communicative abilities required in real-world academic settings evolve due to AI, while test constructs remain anchored to an outdated, unaided-performance model.

The paper categorizes existing literature into three layers to illustrate how the field is responding to AI. The first, and dominant, layer treats AI as operational infrastructure (e.g., automated scoring or proctoring), focusing on efficiency and security rather than construct validity. The second layer views AI as an authentic part of the communicative ecology, acknowledging that academic work is changing. The third layer, which the author identifies as the most critical yet underdeveloped, treats AI as a fundamental challenge to the validity of score interpretations. The author argues that current research often fails to bridge the gap between these layers, leaving a disconnect between how we test and how students actually communicate.

To address this drift, the paper proposes 'bounded AI mediation' as a validity-oriented design principle. Instead of moving toward either total prohibition or unrestricted AI access, this approach suggests a standardized testing condition where all test takers have access to the same institutionally controlled AI assistant. By defining clear assistance boundaries, logging interactions, and designing tasks that distinguish between comprehension support and answer generation, test providers can better align their assessments with the realities of modern academic work. The author concludes that score interpretations must be narrowed and supplemented to accurately reflect a student's readiness for AI-mediated academic environments.