The Rise of AI Homework Cheating
A University of California, Berkeley professor has stopped giving his students graded homework, saying that rampant artificial intelligence use has made it pointless. Renowned mathematician Richard Borcherds stated that it is now nearly impossible to tell the difference between a student-written submission and one created using artificial intelligence. Because of this, he primarily uses in-person exams with no electronic devices to grade his students and evaluate their progress.
Online exams are now a waste of time as several students just use AI to answer the questions. Borcherds noted that average scores increase significantly during online assessments, and grading for correctness creates an incentive for students to rely on automated tools rather than doing the work themselves.
This challenge emerged following the release of early language models. While early iterations were prone to noticeable errors, technology has advanced to the point where distinguishing human-written work from machine-generated text has become exceptionally difficult.
Impact on Academic Evaluations
Various researchers have observed that students routinely achieve higher scores on unmonitored online assessments. While definitive proof of direct AI usage can be difficult to establish in every instance, score discrepancies have led educators to conclude that remote testing no longer accurately reflects course comprehension.
A Brown University professor observed a similar trend when students averaged significantly higher scores on take-home midterms compared to historical averages. Subsequent in-person final exams revealed a sharp drop in correlated high scores, highlighting the disparity between remote and monitored testing environments.
While certain coding assistants have occasionally encouraged users to write their own code rather than generating complete solutions, the sheer volume of available tools makes preventing academic dishonesty exceptionally difficult.
UC Berkeley has evaluated various detection software options but determined they lack sufficient reliability. Consequently, the institution has granted professors autonomy regarding their individual course policies, provided guidelines are clearly stated in syllabi. Borcherds maintains that if the primary disruption caused by automated tools is forcing a return to traditional grading methods, the adjustment remains manageable.
