Building Trustworthy AI in K-12 Education: Insights Into Data Generation and Reinforcement Learning

• In Event: Virtual Poster Session
  In Virtual iPresentation Only: SIG-Technology, Instruction, Cognition and Learning Virtual Poster Session

Abstract

Trustworthy AI has become crucial in various sectors, including education, where K-12 students are highly vulnerable to the influences of AI-generated content. This study explores the application of Reinforcement Learning from Reverse Generated Data (RLRGD) to train AI models for educational purposes. The focus is on creating AI systems that provide accurate, contextually relevant, and educationally enriching responses. We utilized the Algebra Nation Forum Interaction Dataset, employing methods such as simulation-based and reverse data generation. Our results demonstrate that AI models trained with these methods not only meet the standards of trustworthy AI but also enhance student learning by emulating teacher-like responses.

Authors

• Fan Zhang, University of Florida

• Wanli Xing, University of Florida