Enhancing the Safety of Conversational Artificial Intelligence to Support Online Learning Using Educational Big Data (Poster 53)

• In Event: AERA Poster Session Two
  In Poster Session: Collaboration and Peer Interactions

Abstract

Online discussion forums are important pedagogical and social platforms in online learning. However, online discussion forums have been plagued by insufficient interactions among students. A promising approach to automatically support students' online discussions through an organic integration to existing learning environments is conversational artificial intelligence (ConvAI). Few educational studies have examined the safety of ConvAI. This study aimed to investigate and address the safety challenges of ConvAI constructed with educational big data (2,097,139 post-reply pairs by 71,918 students) to support online learning. The results showed that (1) our proposed model could effectively enhance the safety of generated texts and (2) supportive texts could be unsafe, yielding potential threats to creating an engaging and constructive learning environment.

Authors

• Chenglu Li, University of Utah
  Presenting Author

• Wanli Xing, University of Florida
  Non-Presenting Author

• Walter L. Leite, University of Florida
  Non-Presenting Author