Beyond the Bubble Sheet: How Generative AI May Finally Liberate Assessment from Multiple-Choice Testing

Abstract

Multiple-choice questions (MCQs) have dominated educational assessment for decades, particularly in large-scale and high-stakes examinations. Their popularity rests on perceived efficiency, reliability, and objectivity. However, extensive research has highlighted their serious limitations, including construct underrepresentation, susceptibility to guessing, shallow measurement of understanding, and negative washback on learning. This article argues that recent advances in generative artificial intelligence (GenAI), such as large language models, create a realistic opportunity to move beyond MCQs toward open-ended, constructed-response assessment at scale. GenAI systems have the ability to preserve administrative efficiency by automating the evaluation of student-generated answers while significantly increasing validity and insight into student thinking. In this article, we discuss why MCQs remain deeply entrenched in higher education, the pedagogical and epistemic costs of this dependence, and how GenAI-enabled assessment is likely to transform the nature of student preparation and the assessment of learning. The article concludes by describing the circumstances under which GenAI-based assessment can be responsibly implemented and the issues that remain to be worked through.