Cognitive load in mathematics assessment: An exploratory study comparing manual, AI-automated, and human-machine collaboration

Assessing open-ended mathematical tasks imposes significant cognitive demands on teachers, especially in large-scale scoring contexts. As artificial intelligence (AI) tools become more prevalent in educational contexts, it is essential to understand how different assessment approaches influence teachers’ cognitive load and the quality of their evaluative judgments. This exploratory study compared manual, AI-automated, and human-machine collaborative assessment approaches using 300 simulated junior secondary mathematics responses and in-depth interviews with six teachers. Preliminary findings from this sample suggest that performance patterns in the human-machine collaborative approach were consistent with a more favourable balance between extraneous and germane cognitive load, showing the highest overall accuracy and maintaining consistent performance across increasing task volumes. AI-automated scoring performed well on procedural errors but showed marked limitations on conceptually demanding errors. Novice teachers appeared to show greater relative improvements from collaboration in this sample, with their accuracy approaching that of expert teachers working manually. These preliminary findings provide tentative support for the application of Cognitive Load Theory to technology-mediated mathematics assessment and offer initial insights for integrating AI into teacher professional development and large-scale assessment systems, particularly in high-stakes examination contexts.