Leveraging Generative AI to Scale Item Generation Amid Rising Assessment Demands in Education

In the last few years assessment bodies have witnessed larger student numbers, reduced resources and increasing costs of exam administration which have led to the emergence of e-assessment as efficient and cost-effective method of assessment in education. As educational institutions shift towards the complete adoption of e-assessments for high-stake examinations, there are associated challenges that accompany this development. The transition from traditional paper-based examinations to digital formats present both opportunities and challenges for educational systems worldwide. A significant challenge is to develop many test items to support e-assessment. Leveraging technology, such as automated item generation (AIG), can scale up the number of items required in e-assessment but with its fair share of challenges. This paper investigates the computational techniques of scaling up item development to support e-assessment and facilitating a seamless transition to e-assessments while maintaining quality and fairness of assessment. Through a review of literature, the paper explores the challenge of traditional item writing in transitioning to digital e-assessment platforms. Additionally, the paper examined the use of automated item generation tools that are currently in use and their contribution in assessment practice. From the literature, it was observed that template-based AIG has demonstrated great success with many practical use cases. However, despite the success of template-based AIG over traditional item writing, it still requires subject matter experts to develop cognitive models and item models. Finally, the potential of generative artificial intelligence (AI) in addressing this problem was examined. Finally, we explored ChatGPT generative AI model in generating test items, with the objective of finding out its capabilities. By using ChatGPT, it is possible to generate different types of items that are comparable to those created by human authors and this novel development is expected to help assessment providers seeking to navigate the complexities of transitioning to digital high-stakes examinations.