AI bot ChatGPT writes smart essays — should professors worry? (nature.com)

For a long time now, I’ve wanted to study Cold War-era fiction from the communist countries of the era. It’s fairly well-known that authors in those countries were restricted in how they wrote, essentially conscripted as propagandists for the state. As such, from the accounts I’ve read, most stories from those countries are stultifying in their adherence to formulas, protagonists invariably coming to sing the praises of communism (if they weren’t already at the start) after dealing with relatively inconsequential difficulties to give the stories the barest sense of intrigue. By all accounts, nearly any modern reader with even the slightest breadth of experience reading Western fiction would be quickly bored by the lack of variety in these stories (to say nothing of the quality of the prose).

To an extent, I think I undertook a similar study when I read a few romance novels while preparing to write my first novel over a decade ago, familiarizing myself with the conventions of the genre in case I wanted to borrow any of them for my own writing. Romance novels are far from my cup of tea, and it didn’t take much research on my part to understand how the formulas there worked, but I still think I gained a lot from the experience of having read those novels, even if I never write in that genre. In addition to just figuring out the restrictions that Cold War-era communist authors were working under, I’m sure that a more in-depth study would also yield some insights as to how authors who bristled against those communist regimes tried to subvert the formulas in order to rebel against the authoritarianism of their countries, especially as it pertained to the restrictions that writers and other artists were forced to work under.

It was impossible to avoid noticing the attention being paid to the new wave of AI-fueled art bots earlier this year, especially as the output from those bots cluttered up memespace this past summer. On the one hand, I hesitate to dismiss the potential value of these bots, not as replacements for artists but as tools for artists to assist in their development processes. For example, the next time I commission art for my Twitch channel, I think there might be some value in using AI to generate a variety of poses and clothing changes and such, so I can go to an artist with a clearer idea of exactly what I want (and not bother them with lots of small changes throughout the drawing process as I try to figure out just what I want). Unfortunately, the epidemic of existing artists having their art essentially pilfered by these bots, without compensation or even credit, makes the use of them far too problematic for me to even consider, and until these problems are addressed, I won’t be using those art bots for any purpose.

The ChatGPT chat bot had been in the periphery of my news feeds this past week or so, and while I had some desire to look deeper into it (I’ve been fascinated with computers’ attempts to replicate natural speech patterns ever since I played Joe Montana II Sports Talk Football for the Sega Genesis back in high school), I didn’t have the time to do so, what with the end of the semester so close here. When one of those news stories mentioned the possibility of students using the bot to write academic essays, though, learning about ChatGPT immediately became a professional obligation, and after reading a couple of articles, I made sure to set aside some time this weekend to try ChatGPT out and see what it was capable of, especially with my students about to submit their final papers for the term.

I’d like to think that I do a better job of detecting plagiarism than most English teachers, if only because my background in creative writing makes me more sensitive to writing style and tone; writing styles are just like fingerprints, and in addition to finding plagiarized papers online, I’ve been able to detect when some students have had friends write their papers for them, just by comparing the styles of two papers turned in by the same student. I wondered how long it would take me to detect a style in the writing that ChatGPT spits out, given how much advancement we’ve seen in consumer applications of AI in just this past year. It turned out that I only needed about ten or fifteen minutes to figure the style out, so now at least I have that knowledge for all the grading I’m about to do this month.

To be clear, ChatGPT is leagues beyond the last chat bots I experimented with a couple of years ago. Not only can it handle much more difficult queries, but there’s enough variance in its outputs that unless you’re specifically looking for patterns, you probably won’t be able to find them, at least unless you chat with it for hours at a time. I don’t know if I’d call it a quantum leap in chat bot technology, but it’s certainly a very significant improvement, and as with all technology, it’s only going to get better over time. ChatGPT can’t really handle the rigors of writing a college-level paper yet, but that time is probably coming much sooner than some of us — especially English teachers — would like to think about.

I’m already noticing myself shifting my thinking about my upcoming classes because of ChatGPT. Even before the COVID-19 pandemic started, I’d moved all my assignments to online submission out of environmental concerns, and the shorter assignments I give my students at the start of a semester are usually a good baseline from which to gauge each student’s writing style. Now that students will have access to ChatGPT at the start of next semester, though, I’m wondering if I should go back to having some in-class writing assignments, at least during the first few weeks of the term, just to make sure I have some examples of each student’s writing that I know weren’t assisted by the use of computers in any way. I might be able to detect the patterns in ChatGPT’s writing today, but that doesn’t mean I’ll be able to detect them after the next improvements in its output algorithms, whenever those come (which may be as I’m typing these words now).

As much talk as there’s been in fiction circles about the restrictions that authors in communist countries were under during the Cold War, it’s important to remember that at the same time, the CIA was using its influence to turn America’s most prestigious collegiate writing programme into a farm for pro-American (read: pro-capitalist) literature. The techniques may have been different, and the variety of literary output here in the west may have been much wider, but formulaic writing is still formulaic writing. (Speaking of pro-capitalist writing, all the examples of fiction I asked OpenGPT for pretty much read like recaps of bad sitcoms from the eighties. This is not a coincidence.) As long as writers are beholden to these formulas, their individuality and uniqueness is erased, which is why I’ve always focused on helping each of my students develop their individual voice as writers, to help them express the things that no other person — and no other thing — could express in the same way. Whatever uses as a tool ChatGPT may have, if we allow it to become a replacement for our own writing then we are sacrificing our essential identities for the sake of yet another convenience. I have a feeling that advocating this position may be one of the steepest hills we English teachers will ever climb, but given how far text-generating bots have come in just the past couple of years, it’s a hill that we need to start climbing now, before we get buried at the bottom by the endless output of ChatGPT text already being produced.