Online discussions have been widely adopted to solve the challenge of passive learning in online learning environments and foster active student engagements. While successful implementation of active learning begins with an accurate assessment of student discussions, the volume and complexity of unstructured textual data in discussion boards pose another challenge to the task. To address this issue, our paper proposes a novel use of Large Language Models (LLMs), like ChatGPT, to automatically analyze and categorize online discussions based on cognitive engagement levels using the ICAP framework. We measure the effectiveness of this novel technology compared to human assessments and demonstrate its potential to bridge the gap between passive consumption and active learning, revealing a promising step towards leveraging AI technology to improve educational assessment.