Abstract
Higher education has responded to generative AI largely by regulating it by task, specifying where the technology may or may not be used in each assessment. This paper argues that task-based regulation uses the wrong unit of analysis. A single task typically spans several types of cognitive work, and a tool that can be welcomed for one type may quietly displace another. This conceptual paper proposes a four-level framework of cognitive engagement, adapted from classical Chinese expressions about knowing and articulated in contemporary cognitive terms: broad knowledge acquisition, analogical transfer, contextual inference, and original creation. The levels are ordered along a single axis, from cognitive work that is codifiable and reducible to rules and patterns, to work that is situated and dependent on human judgement. This axis maps closely onto the boundary of what generative AI can and cannot do: capability is near-complete at the codifiable end and degrades sharply as work becomes situated. The framework thus functions as a diagnostic lens for design, allowing educators to assign AI a role matched to the cognitive work a task is meant to develop. Two illustrative cases show the framework guiding differentiated, defensible decisions about AI integration within a single assessment.
Keywords
Artificial intelligence, cognitive frameworks, assessment design, academic integrity, generative AI, conceptual framework.
