Structural Modeling of Cognitive Load Reduction in ChatGPT-Supported Learning

Source abstract

The rapid adoption of generative artificial intelligence, particularly large language models such as ChatGPT, is transforming computer-assisted learning and teaching in higher education. This study investigates how ChatGPT can support learning by reducing cognitive load, with particular emphasis on the roles of user technological confidence and affective trust in artificial intelligence within the framework of Cognitive Load Theory. The study employs quantitative analysis based on data from 12,085 respondents across 135 countries, including undergraduate and postgraduate students and academic researchers. The data were analyzed using a PLS-SEM approach implemented in SmartPLS 4.0 with bootstrapping procedures. The results indicate that ChatGPT use is associated with reduced cognitive load and increased user technological confidence. User technological confidence mediates this relationship, while affective trust in AI acts as a moderating factor. A constraining effect was observed, whereby very high levels of trust in AI are associated with more passive interaction with the technology, weakening the overall effect of cognitive load reduction. These findings contribute to a better understanding of how large language models can be effectively and responsibly integrated into higher education, demonstrating that successful human–AI collaboration depends not only on the availability of AI tools but also on fostering appropriate technological confidence and balanced trust to achieve meaningful learning outcomes.