Abstract

Generative artificial intelligence is rapidly diffusing across higher education, yet debates remain polarized between claims that it improves learning and concerns that it weakens independent reasoning. This paper examines how generative AI influences critical thinking in higher education through a PRISMA-guided systematic review (2020–2025), based on records retrieved from Scopus and Web of Science. Using a concept-centric synthesis, evidence is organized around interaction design levers, cognitive mechanisms, and critical thinking outcomes. A Crutch–Coach explanatory lens is advanced to distinguish two ideal-type pathways. In a cognitive crutch pathway, answer-first, product-oriented use is associated with cognitive offloading and overreliance, reduced epistemic effort, and weaker verification practices, with downstream risks for argument quality, source evaluation, and transfer beyond tool use. In a cognitive coach pathway, process-first designs incorporating Socratic questioning, counterargument routines, verification prompts, and reflective checkpoints are more consistently aligned with strengthened epistemic vigilance, improved calibration, and higher-quality justification and error detection. Across studies, key mechanisms (e.g., automation bias, calibration, verification behavior) are rarely measured directly, limiting cumulative comparison. Contributions include an integrative framework, conceptual synthesis matrix, and a research agenda prioritizing mechanism measurement, transfer tests, and boundary-condition analyses across task stakes, assessment design, and student AI literacy.

Download

Share

COinS