Scholarlink

As generative AI tools like ChatGPT are integrated into education, concerns persist about over-reliance, where students seek direct answers without engaging in critical thinking. To address this, we propose VETTING, a framework for designing pedagogically-aligned applications of large language models (LLMs) in education. To explore the practical implementation of this framework, we present a case study examining the impact of a VETTING-informed chatbot that includes a verification layer designed to restrict the provision of direct problem solutions to students. In a randomized trial with 41 undergraduate students, participants interacted with either a general GPT-4o model or a VETTING-informed version designed to promote problem-solving. Findings indicate that students engaged with AI in unexpected ways, rarely asking direct questions. Instead, they sought clarification or broke problems into smaller parts. They treated the chatbot as an extension of their learning materials and rarely followed up when direct answers were unavailable. While students in the VETTING group interacted more frequently, their engagement lacked sustained inquiry. These findings highlight the complexities of designing AI tools that foster deeper learning, but offer insights into how frameworks like VETTING can be implemented to promote targeted pedagogical practices.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

VETTING AI for Deeper Learning: Constraining LLMs to Encourage Student Inquiry

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shan Zhang,
  • Hongming Li,
  • Seiyon M. Lee,
  • Noah L. Schroeder,
  • Anthony F. Botelho

摘要

As generative AI tools like ChatGPT are integrated into education, concerns persist about over-reliance, where students seek direct answers without engaging in critical thinking. To address this, we propose VETTING, a framework for designing pedagogically-aligned applications of large language models (LLMs) in education. To explore the practical implementation of this framework, we present a case study examining the impact of a VETTING-informed chatbot that includes a verification layer designed to restrict the provision of direct problem solutions to students. In a randomized trial with 41 undergraduate students, participants interacted with either a general GPT-4o model or a VETTING-informed version designed to promote problem-solving. Findings indicate that students engaged with AI in unexpected ways, rarely asking direct questions. Instead, they sought clarification or broke problems into smaller parts. They treated the chatbot as an extension of their learning materials and rarely followed up when direct answers were unavailable. While students in the VETTING group interacted more frequently, their engagement lacked sustained inquiry. These findings highlight the complexities of designing AI tools that foster deeper learning, but offer insights into how frameworks like VETTING can be implemented to promote targeted pedagogical practices.