AI in Evaluating Teacher Education Portfolios: A Validity-Centred Framework

Abstract

Professional portfolios act as high-stakes gatekeeping instruments in teacher education, yet their assessment is vulnerable to rater effects, thin feedback, limited scalability and cultural and gender bias. Artificial intelligence (AI) is increasingly proposed as a remedy. This conceptual article argues that AI can enhance portfolio evaluation only when its introduction is governed by a validated framework that places construct validity, fairness, transparency and accountability ahead of efficiency. Using a critical, argument-based analysis of the assessment, validity and AI-in-education literature, the article (a) diagnoses the problems that create demand for AI, (b) analyses the potential and the validity, ethical and practical challenges of AI-supported evaluation, and (c) reviews validated design responses, including structural redesign, triangulation, assessor AI literacy, ethical governance, humanAI collaboration, comparative judgement and aberrant-response detection. It then proposes a six-indicator evaluation framework (validity, fairness, transparency, accountability, scalability and ethical use) that synthesises Messick's unified validity theory, Kane's argument-based validation, Weidlich's Validity Reasoning Matrix and Quality Assurance Agency (QAA) principles. The article concludes that the central question is not whether AI will feature in portfolio assessment but whether it will be used with the critical scrutiny that professional certification demands, and it sets out implications for practice, policy, leadership and research.