Which inference is at risk? Assessment validity reasoning and generative AI

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Generative artificial intelligence has unsettled higher education assessment because it changes what student performances can reasonably be taken to mean. This conceptual article further develops a validity-centred account of this disruption, distinguishing several problems that are often collapsed in debate: construct underrepresentation, construct-irrelevant variance, attribution of performance, conditional extrapolation, and unsupported score use. It then proposes a practical reasoning sequence for assessment redesign that begins with specifying the intended claim and the AI conditions, proceeds through threat diagnosis and proportionate design response, and ends with local evidence. A compact Validity Reasoning Matrix illustrates how course and program teams can align concerns, threatened inferences, redesigns, and evidence needs. The contribution is both conceptual and practical: it offers a language for moving beyond detection and generic AI-proofing and a usable structure for defending assessment claims in AI-inclusive higher education.