A photo of the insides of a computer


I Tried to Teach ChatGPT to Flyfish. It was a Disaster

A review of OpenAI's ChatGPT

As an English professor, I have always welcomed new communications technologies and writing tools in my classes. When blogs became a thing, I replaced papers in my class with blog posts, and the students did all their writing in that medium. Back when Twitter was still fresh and limited to 140 characters, in the salad days of social media, we used a common hashtag in one of my American literature courses to keep an ongoing, public record of our pithy reading notes. (E.g., “Throughout the novel Melville keeps reminding us that we are *reading*, not whaling or sailing. We are navigating/hunting a *text*. #rh2sp13”)

I will admit that I had not paid much attention to the initial brouhaha about OpenAI’s ChatGPT last fall, because at that time I was busy designing a new course. But friends and colleagues kept asking me how I felt about it, as someone who teaches writing; so I realized I should just invite ChatGPT right into my new class. This was an English class, but it was centered on flyfishing. And we weren’t just going to read and write about this practice; the students were going to learn to do it. (It was a two-week intersession class at my university.)

Unfortunately, ChatGPT really underperformed in this context. That’s a vast understatement, actually. ChatGPT barely made it through the first day, when I handed around lengths of monofilament in 15-pound-test and 8-pound-test, and taught the class to tie the two together in a surgeon’s knot. All my human students wrestled with the thin lines and fumbled through the wrapping and looping, some of them getting it. But the whole time, ChatGPT didn’t even pick up the lines. I had deliberately angled my laptop toward me as I demonstrated, assuming that the AI was smart enough to utilize the little spy camera on the top of my screen. But it refused to even try tying knots. Even when we moved on to the improved clinch knot, which is slightly easier, ChatGPT didn’t budge. Presumably it would have written a thesis statement about knots for me, but it refused to get its digits dirty. I know I’m judging the AI unfairly, based on skills outside its intended use; but by now don’t we believe in the dynamism of digital technologies, their potential to exceed our mere human expectations and do things we never expected?

The second day, when we went outside in a wide-open field in City Park and assembled our rods and reels, and threaded the line through the small guides, it only got worse. My computer sat there sullen in the sun, as the rest of the students fitted the four-piece rods together, getting the feel for how to make sure the ferrules were aligned and snug. They had all watched some casting demonstrations online (surely the AI had done that part?), and now it was time to make it real and practice casting on dry land. I went around and tied a bit of yarn to the end of everyone’s leader—but still ChatGPT’s rod was laying there unassembled in front of my keyboard. I kind of wanted to take it aside and say, What the fuck, man?

We stood in a wide circle and all my human students started trying what they’d seen and read about in their class textbooks, as well as what we’d discussed in theory. Casting a fly line is tricky and requires finesse, patience, and a burst of concentrated force at just the right moment. I worked my way around the circle, coaching and giving tips, and sometimes getting behind my students and physically taking their fly rod in my hand, their hand beneath mine, my back almost pressed against theirs—being careful not to be inappropriate, asking permission beforehand—and showing them how a proper cast felt. There is a critical moment in the back cast, or when the rod is up and behind, at about 1:00, when you have to wait an extra second, far longer than it feels like it should take, for the line to “load” in a big loop behind you; and that is the moment when you start the forward cast, “punching” the rod forward with your forearm without slapping it down too abruptly in front of you. It’s something that you have to feel artificially, almost, in order to then get it on your own. I would have thought the AI would have mastered this part, because it’s so counterintuitive.

But no, as I made my way from student to student, the computer continued to sit there—it even went to sleep. At one point I attempted to get behind the screen, after waking it up, and holding the rod handle for ChatGPT, trying to get it to engage, try, goddammit. Just try! But I had to use all my pedagogical restraint to let this student learn on their own terms; maybe they were just observing today, and would surprise the rest of us the next day.

These were all-day-long classes, by the way. My human students got tired and needed breaks. Here again, I thought ChatGPT would have a leg up, what with its AI and all. But it seemed more exhausted and overwhelmed than all the other students combined.

Our course proceeded smoothly otherwise, luckily with pleasant weather and the students each improving in their casting and line handling. So much of flyfishing is simply a matter of controlling the thick fly line that is constantly piling up beneath you, then unfurling in front of you, getting caught in grass or tree limbs…it’s an improvisational choreography of the hands, the long fly rod, and the line itself. Needless to say, ChatGPT sat this skill out, too.

I was getting frustrated with this student’s recalcitrance, but halfway through our course, when we were actually fishing, we had a lesson that I thought the AI would like: reading the water. This is a matter of looking at what is happening on the surface, just below the surface, at the bottom, and around the water. It’s about making sense of various topographical and wetland details, from insect populations to currents to water quality to different fish species present at any given moment. I assumed, given the rich data set that ChatGPT most likely had to work with, that our AI student could produce brilliant—or at least legible—readings of the water before us. But I was wrong again: instead, where the students ventured hypotheses and observations about tidal movements, clarity, and resident prey, ChatGPT stayed silent.

Now I should add that I built in a stipulation to my class, that if any students just didn’t get into the actual physical practice of flyfishing, they could always opt to sit back, reflect, and take notes on what was happening in our class on any given day. And indeed, when some of the human students got worn out or baked by the sun, they would pause for a while and read, watch, and jot down notes about our class. This seemed like a perfectly fine alternative educational path. And so I held out some hope that maybe ChatGPT was soaking it all up, and that in the end its written synthesis would flabbergast us, and we’d realize that the AI had been part of our learning community all along.

The class was to culminate in a collaboratively written essay, in which we would bring together our epiphanies and breakthroughs, in one combined piece. Here then was the chance for ChatGPT to shine: it was all text, now.

I set up the shared Google Doc, invited everyone to join as editors, and the writing and discussion took off. This is another of the new media tools I have used increasingly over the years, in lieu of solitary word processing to produce single-author papers. Google Docs is hardly a perfect system, and it has some maddening aspects, but it is still a professional platform that benefits our students, when they learn to use it well and in its full collaborative mode.

But here once more I completely missed the mark, with respect to our AI classmate. Not only would ChatGPT not be able to sign on to the Google Doc, because it didn’t have a gmail account—it was unclear whether this was on principle or for some more technical reason—but it seemed not even to be able to make sense of what we were doing as a group of thinkers who were writing together about an experience we had just had together. It was as if it hadn’t even been there with us, the past two weeks! I was really shocked at the utter lack of engagement, almost apathy, that the AI displayed just as we were wrapping up our class in the form of writing—the very thing at which ChatGPT was supposed to excel.

I asked my students to think back on the two weeks and derive some lessons. One student, Marie, wrote “At first, I was terrible at flyfishing because my form was off: I would get my foot tangled in the line, or my fly would get stuck in the trees or in rocks in the water. I improved as the class progressed, and began tying my own flies onto my line, and getting my cast to soar gracefully in the air before letting it land nicely on the surface of the water. I learned that patience is key, and it’s okay to take things slow.” This was exactly the kind of lesson I wish ChatGPT would have learned! But it can only be learned by doing it—which the AI preferred not to, if not exactly Bartleby-like.

chat gpt prefers not to flyfish

Another student, Chloe, put it this way: “Flyfishing requires some level of mental and emotional balance as well as physical skill. I found it was so easy to get frustrated or bored, and that’s when the casts got sloppy and when my line would get tangled. Learning to flyfish, standing outside practicing for hours, has almost been meditative in the way that it has forced me to learn how to calm myself and to breathe, especially if I’m upset or distracted. This class wasn’t just learning the basics of flyfishing. It was also learning the mentality surrounding flyfishing.” Maybe this is what was so scary to ChatGPT: the prerequisite of mentality. I don’t want to denigrate the AI, because I really did give them every opportunity to be physically as well as mentally present with us. But somehow, they never quite crossed the threshold, or grappled with that balance that Chloe put her finger on.

My student Kenny went on an interesting tangent about art and perception: “Although I am not an artist, flyfishing is such an active engagement with a natural ecosystem that it feels as though I am the artist in the moment as they soak their surroundings in before painting. I feel the mugginess, I feel the damned mosquitoes, and I interpret the surroundings; I will have different emotions about the mood of the environment, whereas one viewing a painting may only see the emotions of the artist and have their own, but it will never be near what is seemingly an immersive experience.” Perhaps it was the immersion that was missing, for ChatGPT—admittedly, I did have to protect the computer from the elements, throughout our class.

Building on Kenny’s aesthetic inquiry, Mae explained how “I am decisive in my choice to not wear a watch or carry my phone while flyfishing because the constraints of time are not necessary, perhaps they even taint the art form. Chloe said today we must be kind to ourselves while fishing. I think the attitude we display while fishing should carry over more into everyday life.” Now I started to wonder if ChatGPT was a little insulted, or felt like they were being trolled by the other students—after all, it couldn’t even exist without its computer clocks humming in the background, and of course it didn’t have the capacity for kindness.

Blake summed up his experience of the course in pithy terms: “Through this class I have found a new passion. Something that I hope to stick with for the rest of my life.” Morgan put it this way: “The class was given the autonomy to try out techniques, and I gained confidence in myself as a flyfisher. This new type of fishing felt like home.” At this point it was really feeling like a sad sci-fi story, with ChatGPT silently fuming on the side about everything they were missing out on, from mortal life: passion, autonomy, home—I’m almost sorry, ChatGPT.

I decided to give ChatGPT a chance to reflect on the class on its own terms at the Open AI site, but sadly the service appears perpetually “at capacity.” Maybe in the end ChatGPT is just as hyperemployed and stressed out as my students typically are, these days. Still, I was hoping the flyfishing class could be a chance for the AI to learn something new, to get outside, to stretch itself in different ways. To impress us with its wit and facility. I was wrong. Perhaps future iterations of the technology will be better able to integrate with the human students, and engage such a hands-on activity. Nevertheless, I cannot give ChatGPT a positive review, due to their failure to succeed—nay, even try—in my class. (Fortunately, my human students learned a lot in the class, and wrote about it. But that’s another essay.)

Christopher Schaberg is Dorothy Harrell Brown Distinguished Professor of English at Loyola University New Orleans and the author of eight books. His latest is Flyfishing.


The FlyFish Journal Mailing List

We respect your time, and only send you the occasional update.