My philosophy colleague Jason Bridges kindly gave permission to share his follow-up reflections about what his own approach will be in light of the power of the most recent version of ChatGPT and, in particular, why he is skeptical about “relying on relationships of trust to regulate AI use–essentially, using an honors system insofar as the classroom imposes any kind of policy limiting use.” Professor Bridges writes:
While I can recognize what is appealing about this approach, I am afraid I’ve become convinced that it isn’t workable in the long run. And because it is unworkable, I fear it is neither ethical nor in the best interests of our students or the College itself.
I thought I might offer my reasons for view, in case they may be of interest. I’ll note this topic has been especially in my mind as I am co-teaching a Philosophy of AI course this fall (we have 80 students signed up!), and as part of my preparation I have been exploring firsthand what current high-end LLMs can do.
The simple reason the ‘trust and honor’ approach is (I think) unworkable is the incredibly strong incentive structure pushing in the other direction–incentives that will only grow in force over the next few years. We are a year or two away from a time when even an ideal student–one who believes what we tell them about the importance of writing for their learning goals, who wants very much to learn, and has personal integrity–will have staring them in the face the following set of considerations:
1. An LLM can produce an A range paper with no effort on my part. [I’ve verified this with the recent ChatGPT5.]
2. It is easy to render the use of an LLM essentially undetectable.
3. Writing, even if worthy, is painful and difficult.
4. Grades matter enormously for my future.
5. Grades matter comparatively, and other students will be using LLMs.
Even if the counter-considerations appealing to their better nature and lofty ideals of learning suffice to fend off this line of thought through an assignment or two, the question will then be how things look after their first B grade. The pressure toward rationalization will become very, very strong.
There is a further issue embedded in all of this: namely, the need for institutions of higher learning to work to retain their credibility during a period where there are myriad forces arrayed against them.
I had occasion to talk with some students at prestigious high schools in the Bay Area this summer, and they reported that a large percentage of their classmates secretly use AI. They assume their teachers suspect this to be so even while the school seems unwilling to acknowledge this reality. This dynamic causes the school to lose respect in their eyes, and, I think, rightly so.
As I myself am convinced that the process of writing (for oneself, naturally) is crucial for learning how to think effectively in the fields I teach (and, by extension, for the kind of thinking that goes into effective citizenship), I cannot go for the alternative of allowing students to “write” with unbridled use of AI.
The upshot is that I’m moving to a blue-book based approach this fall–ironically, for my class on AI. I’m working out with my co-instructor the details now–we may experiment with supplementary take-home paper assignments that are not graded.
Another possibility for the College going forward is something like Exam4, which has long been used by law schools, allowing written exams to be done securely on a laptop.
I think he’s absolutely right that an “honor code” approach is doomed to failure. I myself am switching to a 4-hour in-class essay exam in my Jurisprudence class in the law school, after having used take-home essay exams for decades. We do use Exam4, or something like it, for in-class exams to block Interrnet access.
Leave a Reply