Artificial intelligence for educational measurement: Where is the value for education?

Artificial intelligence (AI) systems are not intrinsically valuable to education, but rather lend their value from contributing to educational goals. However, when it comes to educational measurement, it is often unclear whether and how AI systems help us to achieve our goals. In this paper, we introduce a way of thinking that helps to clarify how the rules and structures governing educational assessments are impacted by AI systems. Based on a conceptual analysis of the literature, we outline three core elements that should be contemplated when integrating AI systems into assessment: the educational measurement context, the prioritization of different facets of assessment validity, and the social contract between student and teacher. We apply our way of thinking to analyze case studies of AI in item construction, assessing written work, and grading assistance. We show how requiring active reflection on educational aims can inform the realization that gains in subsidiary aims such as efficiency do not provide sufficient warrant for making the move toward AI. We hope that this new way of thinking can instigate critical reflection on what we value in education and how assessments can be designed to reflect those values.