Scholarlink

Generative AI (GenAI) has rapidly become a popular educational tool, offering accessible, personalized support to learners. However, this convenience often leads students to excessively trust GenAI-generated content, neglecting to critically evaluate its accuracy. This study investigates students’ overtrust in GenAI through an assignment involving 39 fourth-year computer science students who used GenAI tools to answer complex, reasoning-based questions. Our findings reveal that 84.6% of students failed to detect inaccuracies in AI-generated responses, demonstrating significant over-reliance. Furthermore, students who depended on GenAI tended to perform weaker on related final exam questions, suggesting a possible reduction in deep understanding. We also discuss the evolving regulations at universities regarding GenAI use and highlight specific limitations observed in current GenAI models. To mitigate these risks, we propose solutions emphasizing increased AI literacy education, improved mechanisms for effective warnings about AI reliability, and training in precise prompting techniques. This work underscores the essential need for strengthening critical analysis skills to ensure GenAI remains a supportive rather than a potentially harmful educational resource.

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

A Call for Action: The Eloquence of GenAI Numbs Students’ Critical Thinking

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Elaheh Jafari,
  • Julita Vassileva

摘要

Generative AI (GenAI) has rapidly become a popular educational tool, offering accessible, personalized support to learners. However, this convenience often leads students to excessively trust GenAI-generated content, neglecting to critically evaluate its accuracy. This study investigates students’ overtrust in GenAI through an assignment involving 39 fourth-year computer science students who used GenAI tools to answer complex, reasoning-based questions. Our findings reveal that 84.6% of students failed to detect inaccuracies in AI-generated responses, demonstrating significant over-reliance. Furthermore, students who depended on GenAI tended to perform weaker on related final exam questions, suggesting a possible reduction in deep understanding. We also discuss the evolving regulations at universities regarding GenAI use and highlight specific limitations observed in current GenAI models. To mitigate these risks, we propose solutions emphasizing increased AI literacy education, improved mechanisms for effective warnings about AI reliability, and training in precise prompting techniques. This work underscores the essential need for strengthening critical analysis skills to ensure GenAI remains a supportive rather than a potentially harmful educational resource.