AI News AI资讯 2d ago Updated 1d ago 更新于 1天前 48

Suspecting AI cheating, Ivy League prof ordered in-person final; scores fell 50% 怀疑学生使用AI作弊,常春藤教授要求线下期末考试;成绩下降50%

A significant portion of elite university students, particularly at Brown University, are utilizing generative AI to cheat on high-stakes academic assessments, substituting genuine learning with automated shortcuts. Professor Roberto Serrano identified widespread cheating in his ECON 1170 course after midterm scores averaged 96/100 with forty perfect scores, a stark deviation from historical averages of 65-80%. The scale of deception was revealed when Serrano switched to in-person finals, causin 布朗大学ECON 1170课程期中考试中,86名学生平均得分高达96分,40人满分,远超该课程历史平均水平(65-80分)。 教授Roberto Serrano通过对比发现答案风格异常,并在期末考试中恢复线下监考,导致平均分骤降至48分,证实大规模AI作弊行为。 期中满分学生中有22人在期末考试前退课或未参加,进一步佐证了成绩与真实能力的不匹配及作弊的普遍性。 布朗大学内部调查显示超半数本科生每周有意使用生成式AI,但同时也普遍担忧其对认知能力的负面影响。

75
Hot 热度
65
Quality 质量
60
Impact 影响力

Analysis 深度分析

TL;DR

  • A significant portion of elite university students, particularly at Brown University, are utilizing generative AI to cheat on high-stakes academic assessments, substituting genuine learning with automated shortcuts.
  • Professor Roberto Serrano identified widespread cheating in his ECON 1170 course after midterm scores averaged 96/100 with forty perfect scores, a stark deviation from historical averages of 65-80%.
  • The scale of deception was revealed when Serrano switched to in-person finals, causing the average score to plummet to 48% and leading to mass course drops among top-performing midterm students.
  • Institutional surveys indicate that while majority of students use GenAI weekly, there is growing anxiety regarding its negative impact on cognitive capacity and authentic education.

Why It Matters

This incident serves as a critical case study for higher education institutions grappling with the integration of generative AI, highlighting the urgent need for robust assessment strategies that verify authentic student learning rather than just output quality. It underscores the ethical and pedagogical crisis where convenience and competition drive students to bypass cognitive development, potentially devaluing degrees and eroding academic integrity across elite institutions.

Technical Details

  • Anomaly Detection in Grading: Professor Serrano utilized qualitative analysis of answer styles ("convoluted style") and quantitative benchmarking against historical data (midterm average of 96 vs. historical 65-80) to identify statistical outliers indicative of AI generation.
  • Controlled Experimentation: The professor implemented a controlled variable change by switching from take-home exams to in-person proctored exams, allowing for a direct comparison of student capability without AI assistance.
  • Correlation Analysis: Data showed a strong correlation between high midterm scores and subsequent behavior; specifically, 22 out of 27 students who dropped the course or missed the final had scored perfectly on the suspected-cheated midterm.
  • Survey Data Integration: The narrative incorporates institutional survey results showing 56% of undergraduates and 67% of graduate students at Brown use GenAI tools daily or weekly, providing context on the prevalence of the technology.

Industry Insight

  • Assessment Redesign Imperative: Universities must move away from traditional take-home essays and open-book formats that are easily compromised by LLMs, shifting toward in-person, process-oriented, or oral examinations that assess reasoning rather than just final answers.
  • Ethical Frameworks Over Bans: Institutions should focus on teaching responsible AI usage and critical thinking skills rather than attempting to police tool usage, as the latter is increasingly ineffective against sophisticated generative models.
  • Cognitive Health Monitoring: Educational leaders must address the "cognitive offloading" phenomenon, ensuring that AI adoption does not lead to a decline in fundamental analytical abilities, which poses long-term risks to the quality of future professional talent.

TL;DR

  • 布朗大学ECON 1170课程期中考试中,86名学生平均得分高达96分,40人满分,远超该课程历史平均水平(65-80分)。
  • 教授Roberto Serrano通过对比发现答案风格异常,并在期末考试中恢复线下监考,导致平均分骤降至48分,证实大规模AI作弊行为。
  • 期中满分学生中有22人在期末考试前退课或未参加,进一步佐证了成绩与真实能力的不匹配及作弊的普遍性。
  • 布朗大学内部调查显示超半数本科生每周有意使用生成式AI,但同时也普遍担忧其对认知能力的负面影响。

为什么值得看

这篇文章揭示了精英高校中生成式AI作弊的隐蔽性与规模化程度,为教育界提供了关于AI检测局限性和学术诚信危机的实证案例。它促使从业者反思如何在保持评估严谨性的同时应对AI带来的挑战,以及高校在政策制定上需更加果断地确立人机协作与学术诚信的边界。

技术解析

  • 异常检测机制:教授通过对比历史成绩分布(均值65-80 vs 96)和文本特征(“非常迂回的风格”),识别出由ChatGPT生成的答案模式,而非依赖传统的代码或指纹检测。
  • 对照实验设计:采用“开卷/带回家”与“线下监考”两种评估模式进行对照。期中开卷考试允许无限时间,期末恢复传统监考,通过分数断崖式下跌(96降至48)量化作弊影响。
  • 数据样本规模:涉及86名选课学生,其中27人因预期无法通过线下考试而退课或缺席,样本具有统计学显著性,反映了高智商群体在压力下的优化行为。
  • 校内调查数据:布朗大学教务长报告指出,56%本科生和67%研究生/医学生每周有意使用GenAI工具,显示AI渗透率极高且已成为常态化工具。

行业启示

  • 评估体系重构:传统的开卷或论文式作业极易被AI攻破,教育机构需转向侧重过程评估、口头答辩或高度情境化的实时考核,以验证学生的真实掌握程度。
  • AI素养与伦理教育:学生将AI视为“捷径”而非学习辅助,反映出深层的伦理缺失。高校需加强AI伦理教育,明确界定“增强学习”与“替代思考”的界限,防止认知能力退化。
  • 行政响应滞后风险:尽管教授公开揭露丑闻,但校方反应被描述为“冷淡”。这提示高校管理层需建立更敏捷的AI学术不端应对机制,避免个案演变为系统性信任危机。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Education AI 教育AI Ethics 伦理 Policy 政策