Can AI-generated questions meet the standards of high-stakes assessment?

Abstract

This study compares AI-generated and human-authored GCSE Biology questions using psychometric analysis, teacher review and expert examiner evaluation. It finds that AI can generate credible assessment questions, but performance is less consistent and AI-generated mark schemes present particular challenges. The findings highlight the continuing importance of human expertise, robust quality assurance and evidence-based evaluation when using AI in high-stakes assessment.

Keywords

• Assessment design
• Examiners
• Mark schemes
• Question difficulty
• Reliability
• Validity