Question for zn/zooey

AI can’t write papers. With a paper you phrase the question in such a way that you have to be familiar with the class issues to respond. Plus I deliberately come down hard on methods and approaches and styles that don’t fit what we’re doing. I am very specific about that. AI writing in this context is very easy to recognize, and if you can’t trace it and prove plagiarism, no matter–invariably AI writing is going to do things I openly and clearly and very specifically don’t allow.

I think we’re not too far away from AI being virtually undetectable. I’d give it 1-3 years, especially with all the money and corporate power behind it, racing for the crown. In five years, it will be the norm, a part of the woodwork. It’s a true threat to creatives, especially, and as a writer and artist, I’m worried. Actors/writers/directors feared it enough to go on strike.

Throw in the race for quantum computing, and we’re in for a true revolution/evolution that will bring some good, but, IMO, a lot of bad, primarily for the non-rich. Literally millions of jobs will be lost to AI. In short, this is looking more and more like a Gattaca/Metropolis scenario, but primarily virtual, rather than biological, though the latter will eventually break through as well.

AI is in its infancy. Too much profit out there for it to remain in that stage for long. If they can recreate a Hemingway short story right now, or make a short talk with Nietzsche possible, it won’t be long before students can input their own essay history, with grammar, style, and idiosyncrasies logged and processed, and then spin out whatever they want. But, again, it’s likely to start out along class lines, and go from there. Money will talk long before everyone can walk, etc.

Bumped into this article today on AI, focusing on schools:

(Did a copy and paste from reader’s mode, using Notepad++. It looks better with regular formatting. Broke into two parts:)

AI is forcing teachers to confront an existential question
Molly Roberts

https://www.washingtonpost.com/opinions/2023/12/12/ai-chatgpt-universities-learning/

“Remember what I told you last week? Forget it.”

This is how Marc Watkins starts many a faculty meeting in the University of Mississippi’s department of writing and rhetoric. The self-fashioned AI guru has been tracking the capabilities of the large language models, such as ChatGPT, that are already transforming how his students write and read — in some cases, by doing both for them.

There is no better place to see the promise and the peril of generative artificial intelligence playing out than in academia. And there’s no better place to see how academia is handling the explosion in ChatGPT and its ilk than at Ole Miss.

In the spring, after students went back to campus eager to enlist robots in their essay-writing, Watkins and his colleagues created the Mississippi AI Institute (not to be confused with the Mississippi AI School, a Mississippi State University venture focused on the artificial insemination of cattle).

The hope is that the institute’s work can eventually be used by campuses across the country. For now, a two-day program this past June at Ole Miss might be the only one of its kind to pay teachers a stipend to educate themselves on artificial intelligence: how students are probably using it today, how they could be using it better and what all of that means for their brains.

The only way to describe what these tools have done to the teaching of writing is to borrow a phrase any professor would mark down as a cliché. They have changed everything.

AI is forcing educators to rethink plagiarism guidelines, grading and even lesson plans. But above all, it is demanding that they decide what education is really about — that teachers ask, in short, “What are we here for, anyway?”

ChatGPT has become to generative AI what Kleenex is to tissues. This most mentioned of tools, however, might be the least of teachers’ worries. Boutique services geared toward composing college essays, the very task Watkins and his colleagues are trying to teach, abound.

Some of their names jangle with techno-jargon, while others strive for the poetic, or at least the academic: Wordtune, Elicit, Fermat.

“Help me write,” read the words atop a Google doc equipped with its AI assistant tool, presumably in the voice of whoever is staring at a blank document waiting for words to come. Watkins finds this disturbing in its vagueness. Help me how?

Other technologies are more explicit about what they’re providing. Wordtune offers the opportunity to select a “spice” to add to your paper.

The “rewrite” option can polish a sloppy sentence; the “explain” option can elaborate on a vague one. There’s also “make a joke” (groan-inducing at best) and “statistical fact” (somewhat more useful, if you’re not worried about AI’s documented propensity to hallucinate). “Counterargument” can — well, you get the picture.

Do you write ad copy? White papers? Plain old emails — or dissertations? Lex, another tool, wants to know. Answer that you write op-eds, and it informs you that “with that type of writing, it can be hard to maintain objectivity while presenting a poignant argument, amidst the pressure of constant deadlines.” (Tell me about it.)

Or you can plug in what you’ve got so far and tell the tool to critique it. Dominic Tovar, an Ole Miss freshman pursuing an engineering degree, likes feeding text into the tool and having it tell him what needs fixing: This sentence is incoherent. This paragraph is too wordy. When things get really rough, he can always type “+++,” a command that prompts Lex to generate the next paragraph — but he thinks students should consider that degree of assistance a last resort.

Other services aim narrower.

Perplexity AI “unlocks the power of knowledge with information discovery and sharing.” This, it turns out, means “does research.” Type something into it, and it spits out a comprehensive answer, always sourced and sometimes bulleted. You might say this is just Google on steroids — but really, it is Google with a bibliography.

Caleb Jackson, a 22-year-old junior at Ole Miss studying part-time, is a fan. This way, he doesn’t have to spend hours between night shifts and online classes trawling the internet for sources. Perplexity can find them, and he can get to writing that much sooner.

Speaking of bibliographies, many students have found themselves filled with despair upon realizing they aren’t actually finished with a paper until they have compiled several pages of APA-style citations complete with annotations. No more! Now, a service called Sutori will handle the pesky copy-pasting and formatting for you.

ChatGPT is sort of in a class of its own, because it can be almost anything its users want it to be as long as they possess one essential skill: prompt engineering. This means, basically, manipulating the machine not only into giving you an answer but also into giving you the kind of answer you’re looking for.

“Write a five-paragraph essay on Virginia Woolf’s ‘To the Lighthouse.’” Too generic? Well, how about “Write a five-paragraph essay on the theme of loss in ‘To the Lighthouse’”? Too high-schoolish? “Add some bigger words, please.” The product might not be ready to turn in the moment it is born, fully formed, from ChatGPT’s head. But with enough tweaking — either by the student or by the machine at the student’s demand — chances are the output can muster at least a passing grade.

Larry Wilson, an Air Force veteran back in school at 43, says ChatGPT and image generators such as Dall-E even aid him in creative pursuits. He crafts comic strips and graphic novels. Sometimes, it’s “difficult getting things in your head out.” But with generative AI, he can explain his vision to a system, and it turns that vision into a tangible image or video. If he sketches out a character to the AI, it returns what he calls an “abundance” of actions, utterances and more that he can insert into the opus of the hour.

Which of these uses are okay? Which aren’t? The harnessing of an AI tool to create an annotated bibliography probably doesn’t rankle even librarians the way relying on that same tool to draft a reflection on Virginia Woolf offends the professor of the modern novel. Why? Because that kind of contemplation goes closer to the heart of what education is really about.

Here’s the bottom line: It’s likely impossible to catch kids using AI to cheat. The detection tools lauded at first as universities’ last bulwark against a horde of scribbling machines have fallen out of favor. They do a poor job identifying cheaters where they do exist — and yet somehow often seem to identify them where they don’t.

See, most notoriously, a professor at Texas A&M University at Commerce who threatened to fail his entire class after using ChatGPT to detect whether it had written their essays. Turns out, it didn’t work.

Or look at Vanderbilt University. The college, in announcing its disabling of one such tool, points out that detectors are more likely to flag material written by non-native English speakers. Its bulletin notes that other companies that pounced on the demand for detectors in the spring have given up. Anyway, none of them was ever able to explain how they could distinguish man from machine — perhaps because, when it came down to it, they couldn’t.

At the Ole Miss summer institute, faculty members see for themselves. “My mother is a fish,” one professor plugs into a service called Turnitin. This is the famous five-word chapter of “As I Lay Dying” by William Faulkner, son of Oxford, Miss. — an ingenious shift into the consciousness of a young boy. The result? Ninety-three percent AI generated. (Probably because the sentence is suspiciously simple, but it goes to show that these detection tools don’t yet appreciate modernism.)

Of course, if the machines can’t detect other machines, that doesn’t mean humans can’t try to. Unsurprisingly, there’s a bit of a “know it when you see it” phenomenon with AI-written work in classes taught by teachers who’ve seen hundreds if not thousands of papers by human students. The trouble for these teachers is figuring out how to react when they do believe they see it.

Sarah Campbell, presenting at the summer institute, described a student essay that appeared, as she put it, “written by an alien.” Or written in the year 1950. Or perhaps written in 1950 by an alien. She responded by asking the student to coffee, where she told the student that she had obviously let them down: “You didn’t know how desperate I am to hear your voice.”

This practically trademarkable Good Teaching Moment cuts to the core of the question colleges now face. They can’t really stop students from using AI in class. They might not be able to notice students have done so at all, and when they do think they’ve noticed, they’ll be acting only on suspicion. But maybe teachers can control the ways in which students use AI in class.

Figuring out exactly which ways those ought to be requires educators to determine what they care about in essays — what they are desperate to hear. The purpose of these papers is for students to demonstrate what they’ve learned, from hard facts to compositional know-how, and for teachers to assess how their pupils are progressing. The answer to what teachers want to get from students in their written work depends on what they want to give to students.

“AI is not meant to avoid opportunities to learn through structured assignments and activities.”

 

interesting thread

w

v

Zooey wrote:

I don’t know what zn means by “invariably AI writing is going to do things I openly and clearly and very specifically don’t allow.”

You basically summed it up. You cannot prove a paper is written by AI but you can tell it is. And virtually everything AI writing does that is bad, and doesn’t work as decent writing, is something you can grade down for being poor writing.

Zooey wrote:

One of my classes spotted right away that it had no ideas in it, but I had to point it out to my other class, and explain how the paper didn’t say anything, even if it said nothing pretty well.

That’s a good example. AI is Hallmark level empty. It has no “eyes” for seeing key details and reasoning about them. It is also prone to vague abstractions.

I like what you said about your assignments there, it’s interesting.

You know what might be fun in this thread to add to the discussion? Ask AI a question about the Rams that all of us know is nuanced and important, and then post what AI says. A question like how did Kyren Williams help the Rams improve their game in the second half of the season. People in the thread would see from such a post that no actual person would ever write that way.

Just do it here! It fits the “what does AI sound like as a writer” discussion.

Try an historical one. Ask about the Faulk trade or something like that.