As much as I want all LLM-based generative AI to be ejected into the sun, it's probably not going to happen any time soon. And since I'm a professor at a university, I have to deal with the role of generative AI in education.

In a world where even AI boosters are noting how it harms cognition, I've got to figure out a policy for my students.

But first, I ought to figure out a policy for myself. I do not like asking students to do things I myself would not do. So, to that end, this part one of a (likely) two-part series on developing a classroom and graduate advising policy. I start with myself. What can I promise students?

My Professional Generative AI Code of Conduct

• I will never use generative AI for grading or commenting upon students' work. If I have grading assistants, they will abide by this rule.
• I will never use generative AI detection services, such as TurnitIn, to police students. If I have grading assistants, they will abide by this rule.
• I will never use generative AI to find or produce any of my course materials. There may be one exception: the generation of AI content for the purposes of criticism or comment. Such a case will be explicitly disclosed.
• I will never assign readings that were produced in whole or in part by generative AI without a disclosure.
• I will not increase the workload of my classes or on advisees beyond what has been appropriate over the past 20 years.
• I also will never use generative AI to conduct my research. There may be an exception: generation of content for the purposes of criticism or comment. Such a case will be explicitly disclosed.

Why these rules?

Let me explain some of my reasoning a bit.

Not using genAI to grade seems pretty obvious to me. Part of the deal of education is that students and profs communicate with each other about ideas. Inserting generative AI into the grading process would undermine that. In addition, there's a workload issue here: if university administrators believe that profs can use genAI to short-circuit the grading process, then admins would likely do things like increase class sizes.

Not using genAI detection services is similar to a rule I've had for years and years: I have never used things like TurnitIn to police students for plagiarism. I avoided it from the outset because I found it to be exploitative -- it's a for-profit business built upon the labor of students. I also think such a practice alters the prof-student relationship to one of distrust and paranoia. GenAI detection services replicate those harms.

Not using generative AI for preparing lectures (or, as my later rule states, my research) is a no-brainer for me. First of all, if I want to prohibit students from using genAI, I ought to walk the walk. Second of all, why in the hell would I subject myself to cognitive atrophy in order to save a bit of time? And why would I risk my reputation by presenting or publishing "hallucinated" work?

Note the exception: on occasion, I might use generative AI for the purposes of comment or criticism. I may decide to use it, either with students or as part of my research, in an effort to better understand it. In fact, I did something like this recently when I attempted to write a document with Microsoft. However, I think these moments will be extremely rare, mainly because in addition to the cognitive harms of genAI, there are extreme environmental harms. I cannot think of too many moments when my need to use these tools for comment would outweigh the harms of contributing to the climate disaster.

As for what readings I assign, I initially planned on only assigning research done without generative AI. The good news is that there's a lot of it, since 99.9% of human history hasn't had to deal with this issue. However, as I've noted elsewhere, academic editors are reporting a deluge of submissions, an uptick that likely is due to the use of generative AI by article writers. In addition, as I noted before, I do not think journal policies go far enough in discouraging this. So, I may have to face the fact that I will at some point assign a reading that was partially (or god forbid, fully) written with generative AI. The best I can offer students is a disclosure about this.

Finally, and I think most importantly, I will not increase the workload of students. I suspect some of my university professor colleagues are doing a bit of math, calculating that, since students now have generative tools, we can ask them to write more than in the past. (I'm basing this suspicion on talking to undergrads.) If so, I think that is a terrible mistake. There is no need to increase "work product" just because there are machines that will spit out slop. If anything, the best reaction to generative AI in education is to do the opposite: slow down and reduce workload in favor of deeper thought.

Next up: my policy for students

I am still drafting a generative AI policy for students. To do this, I have to work at two levels. Undergraduates should get a clear policy. Grad students, on the other hand, might need more bespoke codes of conduct. In fact, I am going to experiment with asking grad students to craft their own policies.