The study introduces a new methodology for assessing collaborative writing between humans and generative artificial intelligence (AI), based on learning analytics and the evidence-centered design. Using data from the CoAuthor writing tool, the researchers identify assessment claims related to knowledge-telling, knowledge transformation, and cognitive presence. They then use epistemic network analysis to make inferences about these claims. The findings reveal significant differences in the writing processes of different CoAuthor user groups, suggesting that this method could be an effective approach to assess human-AI collaborative writing.

 

Publication date: 17 Jan 2024
Project Page: arXiv:2401.08964v1
Paper: https://arxiv.org/pdf/2401.08964