You can't hit the breaks when you're going uphill.
After playing around with ChatGPT, I have decided to allow my history students to use AI. Working out ways to stop them from using it will only create more stress and work than it's worth. I have therefore included instructions in the syllabus about how they can and cannot use AI.
And so I set sail on uncharted waters towards the endless horizon, like generations of intrepid explorers before me.
@eharlitzkern@histodons Maybe not the best picture to illustrate your goal, since in this scene, seconds later Sparrows little boat sunk to the floor of the harbor. 😜
@eharlitzkern@histodons
I haven’t figured out the exact prompt language yet, but I’m planning to do a set of assignments in one of my lower-division courses which is basically “get a ChatGPT essay on this topic and then critique it based out what you’ve read and got out of class discussions.”
@tkinias@mori@histodons That's a very good idea. I discussed that kind of assignment in my teacher team, and we realized that not all prompts work for that kind of assignment. I would suggest you play around with ChatGPT, if you haven't already. I ran the prompts I usually use through ChatGPT and it was really interesting to see what it came back with.
@eharlitzkern@mori@histodons
Yeah, I experimented with “write me an essay on X” prompts for this assignment, and the results tended to be pretty boring but basically hitting the main highlights I’d expect: e.g. for the history of the Ottoman Empire in the 19C it consistently talks about military decline, Tanzimat, etc. I think it’s a good way of encapsulating the ‘conventional wisdom’ on a topic—which could be a good jumping-off point for discussion.
@eharlitzkern@mori@histodons
I haven’t tried any potentially ugly topics with this version of GPT, so I haven’t seen any over racism etc., though when I experimented with an earlier version a couple years ago you could get, um, “interesting” results from asking it questions about Africa.
@tkinias@eharlitzkern@mori@histodons Do you plan to introduce them to some of the critical work on AI, e.g., by Timnit Gebru and Emily Bender, or just give them the tool as another thing that's out there?
@dambaras@eharlitzkern@mori@histodons
This particular class is a GE-level survey on the modern Middle East, so no, we’re not going to dig deep into that kind of stuff because it would take us too far astray from the course topic. But since part of what I want them to do in the course is think beyond surface-level ideas about the Middle East, having them critique a simplistic LLM-generated essay can make for a good task.
@tkinias@eharlitzkern@mori@histodons I understand your approach, but it strikes me that teaching students about what “AI” claims to be vs what it is, and what ideas inform it, IS part of GE.
@dambaras@eharlitzkern@mori@histodons
of course! and we’re currently discussing adding an info literacy requirement to the GE core to address that—I just don’t have time to give them a full intro to the literature on this while also teaching the actual history content (in the same way that teaching basic essay writing is part of GE, but if I fully take that on in my history classes there won’t be any room for the history content)
@tkinias@dambaras@mori@histodons I’m with @tkinias on this. New methodologies need to be hashed out but time is limited when it comes to discussions and teaching content. All in due time.
@tkinias@eharlitzkern@mori@histodons I see that. But I don't think it takes all that much time to introduce the issues and encourage students to be aware of them. Consider: would you show displays of the Hottentot Venus, or Native American remains collected and kept under colonialist systems, without explaining the context?
@dambaras@tkinias@eharlitzkern@histodons There could certainly be some value to having a note or side bar about these kinds of contexts. Wouldn’t take too much time or space
@dambaras@eharlitzkern@mori@histodons
well in my classes the colonial context of such things would actually be the primary content of the course—my goal with the LLMs is basically to get students thinking critically about the tech as much as I can in a course that is really devoted to other matters
@tkinias@eharlitzkern@mori@histodons OK, but thinking critically about the tech requires attention not only to its nature/limitations as a bullshit generator, but also to the racist, eugenicist ideologies (the so-called TESCREAL bundle) of the people hyping and profiting from it.
@dambaras@tkinias@mori@histodons It depends on the class. For my survey course, it will be another tool. For my Honors course, it will be a critical discussion in the classroom. I don’t think I will assign any readings.
@tkinias@mori@histodons When I ran my prompts through ChatGPT 3 the results were run of the mill and very bland. The current versions have been “taught” to avoid problematic replies but for someone creative enough, there are always ways around the safeguards. Also, when ChatGPT 3 couldn’t answer one of my questions it told me that it can only talk about topics from before September 2021, which is kind of interesting. ChatGPT 4 is more up to date, I imagine.
@eharlitzkern@mori@histodons
That’s a safeguard that they added after the fact. Some time last winter I asked ChatGPT about the Georgia Senate election and it happily emitted a completely BS story about how Warnock had lost the election, complete with ruminations on how this was would change the balance of power in the Senate. If you asked it about events after its corpus date, it would simply make shit up.
@tkinias@eharlitzkern@histodons Some of our admins here have discouraged us from requiring students to create ChatGPT accounts, given privacy and security concerns (and concerns re: free training for the AI); they suggest that if we want students to critique an AI paper, we should provide it ourselves.
@KJKesselring@eharlitzkern@histodons
yeah, that’s worth thinking about—I would definitely be fine with finding an alternative assignment for any student who was uncomfortable with it—but I have colleagues in other disciplines who are total ChatGPT fankids and are going to be using it a lot, so I suspect that ship may already have sailed
(I won’t actually require any specific LLM for the assignment, so if they want to play with MS’s or whaterver that’s fine too)
@tkinias@KJKesselring@histodons Yeah, I think the ship has sailed. It sounds more like the administrators are trying to buy time on behalf of the university so they can figure out any legal ramifications. The students have already been using AI for years, and the AI companies are scraping the internet for content; that's how they built their machines in the first place, so training happens with or without us.
@eharlitzkern@tkinias@histodons True, the ship may well have sailed for most, but I think this concern arose from faculty who work in the ethics of computing, so it may not be this cynical in origin.
@KJKesselring@eharlitzkern@histodons
There are definitely major ethical issues here, and to be frank I think the world would be a better place if ChatGPT and the like could be made to disappear entirely. My hope is that by having students engage with the tech under the guidance of faculty who recognize that these things are BS generators they will be somewhat innoculated against the marketing hype when they enter the workforce.
@Ruddiman@histodons Sure! Please let me know if you have any comments or feedback. This really is uncharted territory, so anything that can make the policy better is most appreciated.
@eharlitzkern@Ruddiman@histodons How do the first and fourth bullet point both work? My statement is similar, but I'm wondering how an assignment can be written in their own words and also created by generative AI?
@K_Gregz@Ruddiman@histodons I am being a bit facetious in these requirements, tbh. They are a combination of academic misconduct standards and a "wicked problem." This is how I'm thinking:
If the students follow the bullet points, they will need to do a lot of revision (=writing), fact checking (=studying), source attribution (=learn how to cite), and add a comment about AI (=methodology). Put it all together and you get student-centered learning while allowing AI. Hopefully. 😁
@eharlitzkern@Ruddiman@histodons Ah, ok. I'm telling them they can "consult" or "collaborate" with it for things like revision suggestions or outlines but that everything they turn in for a grade needs to be written by them and any source material, including Chat-GPT-generated content needs to be attributed, fact-checked, and cited. We'll see! 🤞
@K_Gregz@Ruddiman@histodons 🤞🏼🤞🏼Our approaches seem to be similar. It will be interesting to see what we think a few months from now when the semester is over and the grades are in.
Add comment