Governing Generative AI in Educational Assessment: Stakeholder Perspectives on Validity, Fairness, and Integrity in Turkey’s High-Stakes Testing
DOI:
https://doi.org/10.69565/jess.v4i2.422Keywords:
Artificial intelligence, generative AI, educational measurement, high-stakes assessment, responsible AI governance, validity and fairness, reflexive thematic analysis, TurkeyAbstract
The fast proliferation of artificial intelligence (AI) specifically generative AI is transforming the measurement of education in various aspects, such as the development of items, the administration of the test, scoring, security, and reporting. Meanwhile, AI poses novel threats to validity, fairness, integrity, privacy, and societal trust- in high stakes settings, where the outcomes of the assessment are associated with consequential decisions. This qualitative research observes the ways the major educational measurement players in Turkey perceive the opportunities and threats of AI and what they regard as the governance conditions under which they feel responsibly adopt AI in assessment. The study employs a multi-stage qualitative design combining the analysis of documents and semi-structured interviews with experts using their reflexive thematic analysis, which produces explanatory thematic structure. The results have yielded five connected themes, namely purpose-first governance of AI in assessment, (2) validity provided by AI based on construct integrity, evidentiary expectation, and documentation, (3) fairness by continuous monitoring to address drift and subgroup effects, (4) integrity and security issues in the generative AI age necessitating redefined inference rules and assessment jobs, and (5) legitimacy, compliance with privacy, and institutional preparedness as pre-requisites to sustainable implementation. The paper concludes that the Turkish assessment systems that can be responsibly AI-powered involve risk tiering within contexts of use, having clear accountability, documentation that is auditory, ongoing monitoring of fairness, integrity-by-design solutions, and privacy-sensitive governance. The results are used to develop Turkey-specific standards of Responsible AI and monitoring indicators of defensible, equitable, and trustworthy AI-enabled assessment.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 Lamarana Balde, Shazia Khurshid

This work is licensed under a Creative Commons Attribution 4.0 International License.
