Examining undergraduate students’ demonstrated and perceived reliance on ChatGPT in collaborative problem solving

Examining students’ perceived and demonstrated reliance on AI can reveal how accurately students understand and monitor their AI use, which is critical for fostering appropriate reliance on AI. However, little research has examined how to evaluate these reliance levels on ChatGPT in collaborative problem solving (CPS), the extent to which they align, and for what reasons students rely on or refrain from using ChatGPT. This case study analysed the reflections and online notes of 48 Chinese undergraduate students participating in ChatGPT-supported CPS. The results show that although about 23% of students perceived over or high reliance on ChatGPT, only 4.34% demonstrated these levels in their online posts. Students relied on ChatGPT for reasons such as helping learn new concepts and knowledge, enhancing learning efficiency, and providing personalized guidance. In contrast, they refrained from using ChatGPT for reasons such as prioritizing the need to develop autonomy and critical thinking skills, lacking the skills to use ChatGPT, and having doubts about its accuracy and ability to handle complex tasks. The findings advance our understanding of how overreliance can be measured, why it occurs, and the extent to which students demonstrate metacognitive awareness of their ChatGPT use.