From AI-Assisted Performance to Independent Thinking: A Framework for Developing Higher-Order Thinking with Generative AI in Higher Education

Abstract

In an AI-supported task, a student may explain a weak comparison once it has been identified, yet fail to notice a similar problem without a reminder. A decision about further help therefore needs evidence about what the student did without being prompted. We examine such decisions in English-medium university argumentation and evidence evaluation. The framework brings together English-language work on scaffolding, learner-generated reasoning, metacognitive calibration, and cognitive offloading. We distinguish help that supplies a reason from help that initiates a check. Readiness to work without the first does not establish readiness to work without the second. Coding examples and boundary cases make evidence about these functions interpretable, while six conditional propositions connect reconstruction and regulation to adaptation and transfer. A proposed experiment gives every group comparable diagnostic practice but varies whether teachers use its results to adjust assistance. This comparison isolates the value of using the evidence from learning through the diagnostic task itself. We relate the account to four GenAI approaches and illustrate it with a hypothetical sociology lesson. Help may be retained or restored when a new task requires it. The framework and coding rules need validation, particularly beyond the language and disciplinary settings considered here.