Karidi, Amir & Roll (2026) — cs.CY (AIED 2026) 📄 Full text (arXiv)
Karidi, Amir & Roll (2026) present one of the largest empirical analyses to date of authentic (rather than lab-based) interactions between college students and generative AI tools. By analyzing interaction logs at scale, they identify distinct patterns: some students use AI as a llm-powered learning companion for explanation and exploration, while others offload cognitive work entirely — copying outputs without comprehension. The work provides much-needed ecological validity to a literature that has largely relied on survey self-reports and controlled experiments. These findings connect directly to student-experience research, showing that the gap between AI 'use' and AI 'learning' is wide and context-dependent. The study offers actionable design implications for ai-higher-ed-bridge-gap: platforms should scaffold metacognitive engagement with AI outputs rather than optimizing for answer correctness alone. The paper also contributes to generative-ai literacy frameworks by mapping what competencies students actually display in uncontrolled settings, informing ai-literacy curriculum design.