Professor Richard Borcherds of UC Berkeley has discontinued graded homework assignments for his courses. He cites the widespread use of AI tools by students as the primary reason, noting that these systems make it impossible to verify if work is original. Borcherds argues that grading AI-generated submissions is an inefficient use of time and resources.
The issue of AI-assisted academic dishonesty gained prominence following the release of OpenAI's GPT-3.5 in November 2022. While early models were easier to identify due to frequent errors, current technology produces work that is difficult to distinguish from human output. Researchers have noted a significant increase in average scores on unmonitored online exams compared to historical data.
Data from Brown University illustrates this trend, where take-home midterm scores reached an average of 96 percent, compared to a historical range of 65 to 80 percent. When students were required to take final exams in person, performance dropped significantly, with only two students maintaining scores within 10 percent of their midterm results.
UC Berkeley evaluated AI-detection software but concluded that such tools are unreliable. Consequently, the university has granted professors the autonomy to establish their own policies regarding AI, provided these rules are clearly outlined in course syllabi. Borcherds maintains that in-person examinations are currently the only viable solution to ensure academic integrity.
