Abstract:

The purpose of assessment is to make warranted inferences about student thinking, not to reward polished products. Generative systems can now produce fluent prose and step-by-step solutions on demand, which means correctness has become a weak and unreliable proxy for understanding. This paper argues for assessments that remain valid in the presence of AI by eliciting evidence of sense making, movement among representations, strategic choice, critique of imperfect drafts, and reflection that supports transfer. These forms of evidence allow teachers to see how students interpret ideas, recognize structure, and make decisions, even when digital tools are available. The argument draws on contemporary validity theory, evidence-centered assessment design, and mathematics education guidance that emphasizes discourse, reasoning, and representation as foundations for meaningful learning. The paper also identifies practical classroom moves that keep the cognitive work with students and cautions against the use of AI-text detection, which is unreliable and raises serious equity concerns. A guiding question anchors the design: if an AI system could complete the task perfectly, what will students still learn by doing it? This question helps ensure that assessment focuses on thinking, not product polish.

Keywords:

Assessment design, student thinking, artificial intelligence, validity, sense making.