Where Does AI Help and Where Does It Harm? A Stage-Specific Diagnostic for Learning

Students are using generative AI to optimize for the grade, not the learning, producing outputs without the cognitive process that those outputs are meant to develop. A large-scale study confirmed this: Unrestricted AI access preserved scores but impaired exams, severing the output-skill link. Institutions default to binary policies because they lack tools to identify where that link holds and where AI has broken it. This problem is not equity-neutral. EDUCAUSE (2026) reports that 48% of students worry that AI erodes critical thinking. However, one in five cites cost barriers to AI access, meaning the risk concentrates among students least-equipped to manage it. Three equity failures compound: distributional (premium AI makes it costless to capture grades without doing the thinking), procedural (policies are made without evidence about how different populations use AI), and recognition (the system rewards the output without recognizing whether learning occurred). This session introduces the Kolb × Bloom AI-Use Diagnostic Matrix, integrating experiential learning stages with cognitive demand levels to classify AI use as scaffolding or substitution. The matrix produces a risk map: substitution concentrates at higher-order levels where the process matters most and where AI output is easiest to produce. The session includes an interactive diagnostic audit during which participants map AI use in their courses and leave with a concrete alternative to a binary AI policy.