I assign homework problems on online courseware to the students in my calculus-based introductory physics classes to give them opportunities to learn. The homework is intended to enrich the learning they gain by doing collaborative lab and problem-solving exercises during our three-hour studio class periods that meet twice a week.
I’ve always figured that students who want to learn will take their homework problems seriously. Students who short-circuit these homework learning opportunities by lifting solutions from cheating services like Chegg are free to do so, of course. But I hope that more often than not they will be held accountable for their lack of physics understanding on my weekly in-class paper quizzes on which 60% of their grade depends.
I found the message that MIT Emeritus Physics Professor David Pritchard brought to my department during last Thursday’s colloquium talk on the evolution of academic dishonesty in turns comforting and challenging. It was comforting because Pritchard demonstrated that cheating on online homework assignments is strongly correlated with poor performance on the final exams in the physics courses he taught at his university. The challenge was his vision that artificial intelligence tools like GPT, which can be used for cheating, can also be used to improve learning.
Pritchard, who developed the online physics homework system Mastering Physics with his son, identified students in classes using Mastering Physics as homework cheaters if they answered their homework problems within one minute of opening each problem with 100% correctness rates. (Cheating on conceptual questions could not be identified this way)
In the classes that Pritchard studied for cheating, students who cheated on their Mastering Physics problems had a distribution of scores on a pre-test using a standardized assessment (Pritchard used the Mechanics Baseline Test) that was identical to that of those students who didn’t cheat. That is, weaker students were not more likely to cheat than stronger students. The bottom line is that, to paraphrase Taylor Swift, “cheaters gonna cheat”.
When Pritchard analyzed the dependence of final exam grades on several factors including midterm exam grades, online homework scores and pre-test scores, he found that the strongest predictor of a poor final exam score was cheating on the Mastering Physics homework problems. So “cheaters gonna cheat”, but eventually they pay the price for it through poor performance on exams.
For those of us who are Chegg-haters (haters gonna hate, right?), one of the most discouraging developments during the pandemic was the sudden jump in Chegg subscriptions. However, that pandemic-era increase has now been matched by a sharp decrease in Chegg subscriptions driven by the advent of ChatGPT. Cheaters are still gonna cheat, but now many of them are cheating with ChatGPT instead of Chegg. And cheating with ChatGPT is just as bad for student learning as cheating with Chegg.
But Pritchard has a vision for how the use of AI in the classroom can improve student learning instead of harming it. The night before a homework assignment is due, have students explain the difficulties they are having with to an AI. The AI can then summarize the student issues to the instructor, who reads the summary before going to teach his class and adjusts his instruction to deal with the student problems.
The overarching conclusion from all of this is that it is unlikely that AI will ever replace the human instructor completely. It is almost certain that human instructors will harness AI to improve their instruction. But until human relationships become unimportant in student learning – and that will never happen – the human instructor will play the leading role in the classroom.
Bridge to Tomorrow 