Abstract:

As Generative Artificial Intelligence (GenAI) becomes increasingly embedded in educational practice, teacher education faces a critical challenge: supporting future educators not only in using AI tools but in critically mediating their outputs. This study addresses the urgent need to move pre-service teachers beyond a surface-level "utility" mindset—where AI is viewed as a mere time-saver—to a "critical agency" mindset. The research investigates whether structured interaction with specialized custom GPTs can support the internalization of higher-order critical thinking practices in pre-service teachers. Critical thinking is conceptualized here as applied professional reasoning, specifically the ability to interrogate AI-generated information and exercise pedagogical judgment.

The study implemented a 12-week mixed-methods longitudinal design structured around two sequenced GPT-based cognitive scaffolds: the Evidence and Bias Lab (EBL), designed to support skeptical verification, and the Argumentation Coach (ACE), designed to support logical synthesis. Grounded in Vygotsky’s Social Development Theory, these specialized GPTs functioned as a "More Knowledgeable Other" to provide the necessary scaffolding to push learners toward higher levels of independent performance. The central hypothesis was that by isolating the sub-skills of verification and argumentation through AI coaching, students would internalize these cognitive habits as "skill residuals" that persist even when the technology is no longer present.

Quantitative and qualitative analyses were conducted to evaluate the intervention. While quantitative results from a standardized multiple-choice test revealed no statistically significant gains—suggesting a ceiling effect among high-achieving participants—analysis of independent writing samples produced without AI assistance revealed a significant improvement in applied critical thinking. Qualitative content analysis further indicated substantial increases in evaluative skepticism and logical coherence.

Together, these findings suggest that strategically sequenced AI-based scaffolds can support the internalization of critical reasoning practices that persist beyond tool use. The study contributes design-oriented evidence to ongoing debates on cognitive offloading, AI-mediated learning, and the role of generative AI as a temporary epistemic scaffold in teacher education. By providing a framework for "fading" digital scaffolds, this research offers a pedagogical pathway for developing the critical mediation skills necessary for the next generation of educators.

Keywords:

Generative AI, critical thinking, AI GPTs.