Collaborative AI-in-the-loop pedagogical conversational agent to enhance social and cognitive presence in cMOOC

While connectivist MOOCs (cMOOCs) emphasize learner autonomy and peer interaction, the lack of structured facilitation often hinders the development of social and cognitive presence. This study investigates how Pedagogical Conversational Agents (PCAs) enhance learners' social and cognitive presence in cMOOC environments. Using a within-subjects quasi-experimental design, 344 learners (primarily in-service teachers, 93.60 %) participated in a cMOOC course over four weeks. The 54 discussion topics generated by learners were randomly assigned to experimental (with-PCA, N = 19) and control (without-PCA, N = 35) conditions. A collaborative AI-in-the-loop design was implemented where AI-generated content underwent human review before publication. Analysis of 5301 discussion posts using the Community of Inquiry framework revealed that PCAs significantly enhanced open communication (p < .001, r=-0.213) and group cohesion (p=.027, r=-0.106) in social presence, while improving higher-order cognitive processes, including integration (p=.011, r=-0.119) and resolution (p=.028, r=-0.145). Direct interaction with PCAs yielded superior outcomes in affective expression (p=.027, r=0.103) and specific communication behaviors compared to co-present modes. The findings demonstrate that PCAs provide compensatory support by strengthening cMOOCs' relatively weak components of cognitive presence (integration and resolution) rather than providing comprehensive intervention; direct interaction modes with PCA yield more benefit for deep exploration and social engagement. This offers evidence-based design strategies for AI agent implementation in online education.