Critical Thinking & Higher-order Skills

The literature reveals a significant and well-documented concern that AI tools, particularly LLMs like ChatGPT, risk undermining the development of critical thinking and higher-order cognitive skills in K-12 education. This concern manifests primarily through 'cognitive offloading' or 'metacognitive laziness'—where students delegate complex reasoning tasks to AI rather than engaging in the effortful thinking required for deep learning. Multiple empirical studies demonstrate that while AI can improve task performance and efficiency (e.g., essay scores, code generation), these gains often come at the expense of genuine understanding, knowledge transfer, and the development of analytical, evaluative, and creative capacities aligned with higher levels of Bloom's Taxonomy. The evidence suggests AI tools frequently provide direct answers rather than scaffolding the reasoning process, potentially creating dependency and limiting students' exposure to productive cognitive struggle.

However, the literature also identifies promising pedagogical approaches that can harness AI while preserving or enhancing critical thinking. These include: deliberate prompt engineering to elicit Socratic questioning rather than direct answers (extraheric AI, Pedagogical Chain-of-Thought); structured frameworks that create uncertainty to trigger analytical reasoning; hybrid human-AI systems where AI provides hints rather than solutions; and explicit instructional designs that position AI as a tool for metacognitive reflection rather than task completion. The key finding is that the impact on critical thinking is highly dependent on implementation design—AI can either replace or enhance higher-order thinking depending on how it is pedagogically integrated. LMICs face particular risks due to limited teacher capacity to design such sophisticated interventions, inadequate infrastructure for nuanced AI deployment, and pressure to use AI for efficiency rather than learning quality.

The literature reveals a significant and well-documented concern that AI tools reduce opportunities for students to develop critical thinking, analysis, evaluation, and creative problem-solving skills. Five papers directly address this risk through empirical studies or systematic reviews, while six others encounter it as a secondary finding or theoretical consideration. The evidence suggests a dual mechanism of harm: first, AI systems enable cognitive offloading where students bypass effortful thinking by accepting AI-generated solutions without question; second, when AI is integrated into educational tasks, it fundamentally changes the cognitive demands, often reducing tasks to surface-level pattern recognition rather than deep analytical engagement. This manifests across multiple educational levels and contexts, from primary school mathematics to university-level writing.

The most robust evidence comes from systematic reviews and mixed-methods studies examining student behavior with AI dialogue systems like ChatGPT. Researchers consistently find that over-reliance on AI correlates with reduced engagement in critical analysis, decreased effort in constructing logical arguments, and diminished capacity for independent problem-solving. However, the literature also reveals important nuances: the impact varies significantly based on how AI is implemented pedagogically, the specific cognitive level targeted (per Bloom's taxonomy), and whether students receive explicit instruction in AI literacy. Notably, several studies demonstrate that AI can support higher-order thinking when used as a scaffolding tool rather than a replacement for cognitive effort, suggesting that pedagogical design is the critical mediating factor.