Fly-Fishing With AI

Last year I wrote a story about how I tried to teach ChatGPT to fly-fish, and how it didn’t work out.1 But it was mostly a story—as in, fiction. Or at least creative nonfiction. I had in fact taught an actual fly-fishing course2 recently at Loyola University New Orleans, which was inspired by a book3 I had recently written. My fly-fishing course took place around the time that a lot of my colleagues and administrators on my campus were getting whipped up about the threats that artificial intelligence (AI) posed to learning: Would there be an unstoppable cheating rampage? Would certain creative skills be abruptly outmoded by these new creaturely computer programs? In short, were we doomed? I had thought this was all rather comical as I was teaching my students to do something with their hands and bodies (casting a fly rod and line) while interacting with the dynamic elements around them, from wind and trees to fish and flying hooks, all within the urban constraints of New Orleans. ChatGPT and the like felt very distant from this space and time.

Over the following months AI morphed and evolved and promised to disrupt ever new domains. I was watching from afar but growing more intrigued. Then my colleague Ian Bogost wrote a piece for The Atlantic arguing that we were getting AI-generated images all wrong: they weren’t an inherent existential threat to visual art or to photographic realism but rather could be “a tool to supercharge your imagination.” He argued that “AI images allow people to visualize a concept or an idea—any concept or idea—in a way previously unimaginable.”4 I found Ian’s article clarifying and encouraging, perhaps in no small part because my 13-year-old son had started experimenting with AI-generated images on his school iPad, showing me comical collages and sometimes strangely profound scenes. It wasn’t just a shortcut for making art, but a new kind of medium, or a synthesizer of sorts.

I should say here that I’d changed jobs since I taught that fly-fishing class.5 My family and I moved 650 miles up the Mississippi River, where I took on a new leadership role at Washington University in St. Louis. It was a dramatic shift away from the relaxed life of an English professor. In New Orleans I taught two days a week; the other days I read, wrote, planned my classes—and fly-fished for an hour each day at the end of Bayou St. John, a few blocks from my home.

In St. Louis I was too busy to fly-fish with any regularity, though I did dash over to Forest Park one day during my lunch break and promptly caught a bunch of gorgeous green sunfish with my small tenkara rod, off a bridge near the golf course. But mostly my fly-fishing had receded into places of memory, or drifted off into future fantasy realms.

Thinking about Ian’s counterintuitive philosophy of AI made me wonder: Could I get an imaginative jolt around fly-fishing from some images generated by artificial intelligence? Might bizarre representations and speculative mashups of fly-fishing—or what an AI produces as fly-fishing—satisfy some part of my aesthetic longing for an activity I was missing in my day-to-day life? If so: Where to start? I searched online for free AI image generators (I was AI-curious, at best), and after being stymied by a few, settled on a friendly mollusk-themed site called Gencraft. I selected this one because it was easy to sign into using my Gmail account, and fairly straightforward to use. I wasn’t trying for the best AI experience, in other words, but for a lower-budget, more democratic version—because this is also how I approach fly-fishing: not as a rarefied or esoteric activity requiring destination travel, but something that can be practiced at minimal cost, and anywhere there’s water.

At Gencraft, I wrote a prompt based on what I was missing, what I wanted to see—namely, my fishing haunts in northern Michigan, where I’d normally have been all summer: An English professor fly-fishing for bass in a glacial lake. Here’s what appeared:

Well that sure isn’t Michigan. The word glacial must have triggered imagistic hits for the national park in Montana, in the brain of the AI. Then there is the shadow double rod, which I’ve discovered is an interesting tic that reappears across AI-generated images of fly-fishing. Maybe it is a misinterpretation of the fly line, understood as a second rod? And what about the lack of a reel, and that odd-angle grip? There’s something otherworldly about it—and possibly that’s what Ian was talking about, the way I have to contort my mind around an object that is uncanny in its almost familiarity. No sign of bass here, but it doesn’t look like their kind of habitat anyway. As for the English professor—is that what we look like? Some kind of generic, Hemingwayesque schlub? That was the first of two attempts. (Gencraft produces two images for each prompt; you get to ten prompts per day, for free.) Here was the second:

This one reminded me of that Studio Ghibli film Pom Poko, when the racoons-qua-humans wipe their faces off to scare the intruding humans. (Did it look like THIS?) On the other hand, the fly line now bears closer resemblance to what it might actually feel like to fly-fish there. And there is no shadow rod here. Still the topographic dissonance, but a pleasant picture to stare deeply into nonetheless. Maybe I need to be more specific.

I added detail to my prompt: An English professor fly-fishing for bass in a glacial pond in the sand dunes next to Lake Michigan.

Well okay, I appreciate what look to be vaguely white pines. I can see the layers of branches they are alluding to, I really can. And the body of water, it actually looks incredibly close to a secret kettle pond that I know about, up in Michigan. It is almost transporting me to the ecosystem I know and love deeply. And no phantom second rod! But the line and the bent rod tip almost make it look like the fisherperson (wait, is that Red Skull?!?) is preparing to bow-and-arrow cast toward the reed-bank. This was close. The second image, not so much:

Who is this lost angler with a broken rod in the midst of an expanseless watershed? The rod is fucked. And maybe this image, warped as it is, hits too close to home for any fly-fisher who has snapped a rod mid-excursion. There will be no fishing anymore today for this poor sod. His face shows that he hasn’t even registered the extent of the disaster yet. And what is that drone-like, batteries-not-included object hovering at butt-level? A crumpled Doritos bag caught in freefall, vibrant litter?6 Those cargo pants are not waders; the fishing vest is actually more like a personal flotation device, the pond more of a slough in a meadow…. Everything is just a little off in this image, most glaringly in the destroyed rod, snapped and mangled by a fish or fate unknown. But did it get my imagination working?

Reader, it did.

I drilled down deeper into the data, modifying my prompt once more, to see if I can approximate my fishing around another location. As I mentioned earlier, I don’t approach fly-fishing as an elitist or Romantic endeavor, but as an everyday practice, in whatever waters are nearby. This would be more evident in how I practiced fly-fishing in my former home, and so I adjusted the prompt accordingly:

An English professor fly-fishing for bass in an urban bayou in New Orleans.

Here are the results:

In perhaps the most surreal image yet, the fishing rod becomes a fish-rod, its handle merging into the tail of a trout-like thing. Note the six fingers, an oft-noted propensity of AI image generation. In the background is something resembling a bayou and shotgun homes on piers. That overhanging live oak is unmistakable. Even the tree trunk in the bottom right corner looks accurately cypress-ish. The ecosystem is there, if offset by the professorial mutant angler and freak device.

The second mashup emits a rising-sea-levels aura—or, just another day in the Anthropocene. I recall a flash flood that did this just a few years ago, the water inundating the city in a matter of hours and totaling hundreds of cars parked on our street. This is not what my New Orleans fly-fishing looked like, although it might be an everyday scene soon, given the realities of climate change. The faceless angler abides. 

It’s possible that my “bayou” descriptor threw off the algorithm. Let me try again: An English professor fly-fishing for bass in an urban canal in New Orleans.

This one surprised me for its utter lack of fishing, fly or otherwise. Just an empty canal, and a lone bystander staring into a storefront, zombie-like. Come to think of it, it looks as though a waterlogged French Quarter has become the set for an episode of The Last of Us (HBO, 2023-present). And again, no fly-fishing! Possibly some gear or netting near the breakwall on the left. Fascinating, if also a bit disturbing. 

I want to get back to Michigan, to a place that brings me more peace. I’ll try one more prompt: A professor of airports fly-fishing for northern pike in a weedy lake in northern Michigan.

These images got my imagination working, yes. They are weird, and inviting, and sometimes even lovely. They fundamentally don’t accurately represent fly-fishing—in the equipment or action—but they do other things, like fuzzy photographs of aliens trying to fit in among us. Or like I have glimpsed other goldilocks planets, as in Interstellar, places that can sustain ecosystems and elements as we know them, even if parts of the lifeworlds are surreal looking or scaled differently.

The day after I co-created these images, I went into Forest Park to a riffle I had been scouting, only to find it barricaded off and heavy machinery excavating the site, reshaping a bend in the artificial river. I parked nearby and walked ‘upstream’ a hundred feet or so, just past the construction fence, to where the water pooled into a body more pond-like. I had brought my two-weight rod and attached two nymphs to my tippet, a beadhead with a bit of pulsing hackle in front, and a wispy little green pupa a foot behind. I cast into the murky pond and brought it back with a slow retrieve, feeling for bites on the drop. It was mid-November, but it had been a warm fall and I had seen fish scooting along the shoreline.

Sure enough, the fish were there. Over the next hour I caught a dozen or so silver-dollar bluegills, throats aglow with yellow, as well as five or six green sunfish, their variegated gill-plate designs as intricate as the backs of brook trout.

After my modest experiment with Gencraft I think I felt the fly-fishing differently, as a more spectral act—something more computer-generated, even. This pond was artificial, but so was the whole experience: progress, curated leisure, my ‘time-off’ of work. Standing by the water’s edge, I had the uncanny sensation of being watched, and when I glanced over my shoulder before my next cast, I noticed someone holding up their phone and taking pictures of me. Probably just a bemused runner, who had never seen someone fly-fishing in Forest Park. Fine. But then I thought, those digital pictures of me—whether texted to a friend or simply stored on a photo reel—will feed the AI algorithms, helping to refine future imagery. This is fine too, I suppose.

I continued to catch fish off a submerged branch, suspecting that I was doing two things simultaneously: one for myself, and another for AI. It was like a twist on that Thoreau line from Walden, when he’s fishing at night and because of the reflection on the lake’s surface, thinks that he might as well be casting both into the water and into the stars: “It was very queer, especially in dark nights, when your thoughts had wandered to vast and cosmogonal themes in other spheres, to feel this faint jerk, which came to interrupt your dreams and link you to Nature again. It seemed as if I might next cast my line upward into the air, as well as downward into this element, which was scarcely more dense. Thus I caught two fishes as it were with one hook.”7

My foray with AI was in a sense a new form of American Romanticism. It had made everything feel generated, somehow projected yet frozen in time and space. Or maybe I’m just imagining these phenomenological effects. Imagination is a strange thing. On the one hand, AI gave me a new way to index my flights of fancy. On the other hand, AI also felt like little more than heaps of digital detritus, piling ever higher into a cloud where precipitation will not accumulate.

Christopher Schaberg is Director of Public Scholarship at Washington University in St. Louis, founding co-editor of the Object Lessons series, and the author of ten books, including Adventure: An Argument for Limits (Bloomsbury, 2023), and Fly-Fishing (Duke University Press, 2023).