C. Thi Nguyen

That’s it! She also returns to the idea in her new book, AI Mirror.

About 30, and I doubt it would work nearly as well much larger. Ironically, we were also reading stuff about how size and scale limits real democratic conversation.

That’s it!

Ironically - or perhaps appropriately - we did this exercise as we were simultaneously reading about open democracy and citizen assemblies and various alternative democratic proposals.

Anyway: here's to my students. They took it seriously, they argued it out, and they came up with something that I doubted at first, but ended admiring more than any system I've designed on my own.

At least, it was better than anything else I've thought of and tried on my own.

And I think... it mostly worked? I mean, I'm sure the system is cheatable to somebody who was trying to break it. But for a system designed to mostly get students actually engaged in the critical act of thinking, without a punitive ChatGPT detection system... it was... really good?

In general the class vibe was: that they knew they were going into a work environment in which AI would be a tempting tool, and that some of them would want to use it in various ways, but that they also wanted to have systems to keep them honest about actually developing real thinking skills.

We decided I should do a default light-commenting, and hopefully anybody that AI-wrote the final paper wouldn't request heavier commenting, and it would save me from heavily commenting on an AI paper.)

(PS, we also stole a thing I've been doing for other reasons as protection for my soul. I have traditionally said I will default give students medium-intensity comments, but students can request lighter commenting or heavier commenting.

Their idea and goal was to keep each other honest about actually understanding the material by demanding repeated presentations to each other in class.... and it mostly worked?

And it was... amazing? Like, the quality of discussion was extremely high, students ended up doing a ton of live thinking and live tweaking and pushing on each other. It wasn't the same as writing, but it most of the students were seriously thinking, and pushing, and applying.

And the third workshop was going to be a presentation of their ideas in outline, and the fourth workshop was a simple draft reading workshop.

The second workshop was one in which they wanted me to break them into working groups based around similar topics, where they would continue to workshop, refine, and throw ideas at each other.

They hammered out a much more complicated set of group workshops, participation in which would constitute the majority of their final project grade. The first workshop was a "brainstorming" workshop, where they mixed up cases they were interested with theories, and helped each other design papers.

I often run, in the past, a kind of traditional workshop structure. They come in with an outline of their paper, and workshop it. Then a draft, then workshop it. Etc.

And then we did their assignment structure they built, and it was fantastic.

Like, way better than anything I've designed on my own.

By the midpoint of the class, I was pretty bitter. It actually started looking like a counterexample to the ideal of inclusive democracy. That maybe I should go back to being the authoritarian technocrat, that my teaching experience was much more important.

Then they presented - and the miserable compromises came. A lot of them hated each others' suggestions and they ended up cobbling together what looked to me like a terrible, misshapen assignment structure of what looked to me like BS-filled group participation assignments, at cross-purposes.

A majority of the groups came up with some kind of process that emphasized live, in-class interactions, and non-AI-usable live interaction. Since I gave them the constraint that I didn't have time to orally examine each one separately, they came up with various group in-class exercises.

The discussion process was amazing. It created some of the best and most engaged conversations in class I've had. My theory is that I *staked* the conversation - I made it important, and put my money where my mouth was by giving them real power, and real consequences.

Two general themes they agreed on immediately:

1. If they used AI for everything, they would learn nothing, and come out of college with no valuable skills, and be replaceable by an AI.

2. All available AI detection software was unreliable and unjust. (I agree with this.)

Students seemed to take it super seriously. A few wanted to abuse the power to design an easy class, but the vast majority seemed invested in thinking about what they wanted out of education, what skills they needed in the new era, and what assignments were *for*.

I told them they needed to give me a ChatGPT policy, and if there was a ban, they needed to give me a method of detection and enforcement.

I also told them if this ended up with me carefully commenting on mostly AI-written papers, I would probably quit the profession.

Qualification: 60% of the class evaluation was based around semi-traditional in-class exams. I was letting them design the "final project" assignment which was about 30% of the class. So: not a complete democracy. Sorry.

And as I was doing all this, I started to think: why haven't I done this before? Isn't it a form of hypocrisy to have a class where I worry about authoritarian systems - and then to impose my own assignment structure without any input from those it rules?

We also did a bunch of work on democratic inclusiveness, and various theories that the point of democracy is that no single authority can account for all the important reasons. That every single participant in a society has a different angle, a different form of relevant knowledge.

This was crucial - it really primed students to be concerned about what parts of their skill - and what parts of their humanity - might erode if they unthinkingly used automation. It set the tone in I think a crucial way.

Vallor's paper really hit the students hard - it was about up-skilling cases where automation replaces a useless skill and gives us more time, vs. cases where automation replaces a crucial skill and de-skills us, makes us fall out of practice. Her core cases are about AI-based moral judgment.

Background: this is for a cross-listed class, that was both Tech and Design Ethics, and Political Philosophy. As part of the run-up, we read a bunch of stuff about the impact of technology and automation on human skill and learning. In particular, we read @shannonvallor.bsky.social on deskilling.