Students are using ChatGPT to do their HW. Here are things I've heard and some of my thoughts on the issue (Lance also added some comments). I have no strong opinions on the issue. Some of what I say here applies to any AI or, for that matter, old-fashion cheating by having your friend do the homework for you or by going to the web for the answer (Is ChatGPT going to the web for the answer but with a much better search tool?)
1) Ban the use of ChatGPT. That might be impossible.
2) Allow them to use ChatGPT but they must take what it outputs and put it in their own words. This should work in an advanced course where the students want to be there and are mature, but might be problematic in, say, Freshman calculus. This is also a problem if you want to see if they can come up with the answer on their own. Lance: How do you know if the words come from AI or the student?
3) Assign problems that ChatGPT does not do well on. I can do this in Ramsey Theorem and (to some extent) in Honors Discrete Math but (a) over time it will be harder, and (b) this may be hard to do in a standard course like calculus. Lance: Pretty much impossible for most undergrad courses. Bill: Slight disagreement- computer science is a fast-moving field where recent papers can be used to get problem sets before ChatGPT has digested the paper. Sometimes.
4) For some assignments have them use ChatGPT and hand in both what chatGPT output and the student's rewriting of it. If ChatGPT made a mistake they should also report that. Lance: Many would just have AI rewrite the AI. Bill: Alas true.
5) Revise the curriculum so that the course is about HOW to use ChatGPT intelligently including cases where it is wrong and types of problems where it is wrong (do we know that?). Lance: And that changes as models change. Bill: True but not necessarily bad.
6) Make sure that a large percent of the grade is based on in-class exams. I don't like this but it may become needed. Lance: I just graded homeworks for completion instead of correctness so that students didn't feel they needed to use AI to keep up with their classmates. Bill: I like the idea but they may still use AI just because they are lazy.
7) The in-class exams will have variants of the homework so if a student used ChatGPT this will test if they actually learned the material (I do this anyway and warn the students on day one that I will do this to discourage cheating of any kind.) Lance: This works until the students wear AI glasses. Bill: I can't tell if you are kidding. More seriously, how far off is that? Lance: Not far.
8) Abolish Grades. Lance thinks yes and blogged about it here. Abstractly I wish students were mature enough to not need grades as a motivation. But even putting that aside, we do need some way to tell if students are good enough for a job or for grad school. Lance: We do need that, but grades aren't playing that role well anymore. Bill: What do do about this is a topic for another blog post.
9) In a small class have the students do the homework by meeting with them and having them tell you the answer, and you can ask follow up questions. Perhaps have them do this in groups. Lance: Panos Ipeirotis used AI to automate oral questioning. Bill: Looks great, though I'll wait till they work out the bugs.
So---readers- what have been your experiences?
It is a fascinating idea to use AI to conduct the exam. Possibly that is the future. I would require the students to be on site to eliminate cheating. Maybe each university will have small AI rooms where you can conduct such exams, but outside exam period use them to learn with the help of AI.
ReplyDeleteRegarding homeworks, in Hungary they never counted significantly towards your grade, because cheating and dishonesty are more widespread. I'm considering that from next semester they won't count at all to the grades, but students would be allowed to submit them optionally to get feedback, with the requirement that the first line should read: "Dear Professor, I humbly beg you to be so kind as to give feedback on my work."