Brown Professor Suspects Majority of His Class Used AI to Cheat
Brown University professor Roberto Serrano identified widespread AI-assisted cheating in a take-home midterm, evidenced by unnatural writing styles and proof methods that mirrored ChatGPT outputs. The professor invalidated the midterm scores after an in-person final exam revealed a drastic performance drop, with the class average falling from 96% to 48.6%. University administrators responded with bureaucratic delays and requests for individual complaints, which Serrano criticized as insufficient
Analysis
TL;DR
- Brown University professor Roberto Serrano identified widespread AI-assisted cheating in a take-home midterm, evidenced by unnatural writing styles and proof methods that mirrored ChatGPT outputs.
- The professor invalidated the midterm scores after an in-person final exam revealed a drastic performance drop, with the class average falling from 96% to 48.6%.
- University administrators responded with bureaucratic delays and requests for individual complaints, which Serrano criticized as insufficient and reliant on flawed AI-detection tools.
- The incident highlights the systemic difficulty institutions face in adjudicating mass cheating cases where traditional academic integrity protocols are overwhelmed by the scale of AI usage.
Why It Matters
This case serves as a critical warning to educational institutions regarding the fragility of remote assessment models in the age of generative AI. It underscores the urgent need for universities to develop scalable, fair, and efficient protocols for handling mass academic integrity violations rather than relying on outdated, resource-intensive individual adjudication processes.
Technical Details
- Detection Methodology: The instructor used qualitative analysis of student responses, specifically noting "convoluted style" and the use of unnecessarily complex mathematical proofs (e.g., contradiction arguments instead of direct arguments) that matched ChatGPT's typical output patterns.
- Statistical Anomaly: The midterm average was 96%, a significant deviation from historical norms of 65-80%, despite the exam being designed to be more challenging due to the open-book format.
- Validation Experiment: The professor administered an in-person final exam to verify suspicions; the resulting average of 48.6% confirmed that the high midterm scores were not indicative of actual student knowledge.
- Administrative Friction: The university’s Standing Committee on the Academic Code required individual complaints for each suspected student, ignoring the statistical evidence of systemic cheating and proposing the use of AI-detection software known for high error rates.
Industry Insight
- Shift in Assessment Design: Institutions must move away from traditional take-home exams for core competency assessments and towards in-person testing or continuous, authentic evaluation methods that are resistant to AI generation.
- Policy Modernization: Academic integrity policies need to be updated to address "mass cheating" scenarios, allowing for aggregate statistical evidence of misconduct rather than requiring impossible-to-meet burdens of proof for individual cases.
- Faculty Support Systems: Universities must provide administrative and legal support to faculty facing large-scale integrity issues, as the current burden of proof and adjudication process disproportionately penalizes educators who attempt to uphold standards.
Disclaimer: The above content is generated by AI and is for reference only.