The Clankers Are Touching Grass

AI discovers birdsong, fishing trips, and mushroom foraging.

AI tends to make the biggest splash when it writes code or generates images or threatens entire industries. But some of its most interesting work is happening much further from the spotlight, in the woods, on the water, and in the field, where a growing collection of apps is trying to help people identify birds, plan fishing trips, and figure out which mushrooms won’t kill them. The results vary in ways that are worth a look.

Shazam for the woods

There’s a moment, if you spend any time with the Merlin Bird ID app, that feels like a magic trick. You hold up your phone, tap the microphone, and birdsong that was previously just pleasant background noise starts resolving into names. The app converts sound into a live spectrogram and tags individual species as they vocalize, even when several are singing at once.

Naturalist Drew Monkman captured the experience nicely last November. On a quiet October morning at a provincial park in Ontario, his ears picked up only the chip notes of yellow-rumped warblers. Merlin, listening alongside him, surfaced white-throated sparrow, golden-crowned kinglet, brown creeper, and then, to his surprise, scarlet tanager. He cupped his ears, listened harder, and there it was: a faint chik-brr he’d never have caught unaided. Binoculars confirmed it.

Merlin is built by the Cornell Lab of Ornithology and powered by eBird, the world’s largest database of bird sightings. It now recognizes over 2,000 species by sound, runs entirely offline, and is free with no subscription. It has more than 10 million downloads and may be the single most successful consumer AI application that nobody in Silicon Valley talks about. The model improves because birders contribute recordings to eBird, which train the next version of the algorithm, which attracts more birders, which generates more data. The broader platform, iNaturalist, hit four million contributing accounts in January, with its observations feeding conservation research across 128 countries.

The key thing about Merlin, and the reason birders love it rather than resenting it, is that it makes one more attentive, not less. The app gives a name; the user still has to listen, look, and confirm. A mockingbird doing impressions will fool it. Background noise generates false positives. But as a training tool for the ear, there’s nothing else like it.

The 5 a.m. guide

Where Merlin identifies what’s already present, fishing AI tries to advise you on conditions you haven’t encountered before.

The most ambitious entry is onWater Fish, whose Angler Intelligence suite launched last October. The Chat feature draws on live river flow gauges, hatch forecasts, weather feeds, access points, and historical fishing patterns, then synthesizes all of it into a plain-language game plan for a specific location and day. There’s also Measure, a vision-AI tool that estimates a fish’s length and weight from a photo, useful for logging catches without overhandling the animal and for contributing size data to conservation partners like Trout Unlimited.

Sports Illustrated’s Kurt Mazurek put the system to a real-world test in December. Visiting unfamiliar water near Dunnellon, Florida, he asked Angler Intelligence to build him a plan for largemouth bass. What the app didn’t know was that Mazurek’s fishing partner that day was veteran Florida bass pro Bernie Schultz, who has fished those waters for decades. Schultz didn’t know the AI had been consulted. The result: both the app and the pro pointed to the same areas near the confluence of the Rainbow and Withlacoochee Rivers, targeting vegetation edges and submerged wood. The app even flagged the unusual clarity of the Rainbow River. It wasn’t perfect, but it landed in roughly the same place a lifelong local would, and it did it in seconds instead of years.

For anyone who’s ever shown up to new water with a boat and no idea where to start, that’s a real service. It won’t teach the feel of a subtle take, and it can’t replace the slow process of actually learning a piece of river. But the old-timers at the fly shop aren’t open at 5 a.m., and sometimes one just needs a reasonable starting point.

The coin flip

Mushroom foraging has boomed in recent years, and AI has taken notice. App stores now carry dozens of fungal identification tools: ShroomID, ForageFinder, Picture Mushroom, and a growing roster of ChatGPT-powered bots. The pitch sounds just like Merlin’s: point your camera, snap a photo, get an answer.

The problem is that mushrooms are not birds. A bird on a branch presents its key features to a camera more or less the way a field guide would. A mushroom does not. Accurate fungal identification often depends on the underside of the cap, the base of the stem, the substrate it’s growing on, and critically, its smell. (Mushrooms can smell like maple syrup, coconut, almond extract, dead fish, or dirty socks, and those differences matter for identification.) An app working from a single photograph is missing most of the relevant information.

A 2022 study by Australian poison researchers tested three popular mushroom apps against photographs of 78 species and found the best performer, Picture Mushroom, correctly identified mushrooms only about half the time. Its accuracy on toxic species dropped to 44 percent. That’s a coin flip with consequences, and they aren’t theoretical. Public Citizen’s 2024 report documented hospitalizations from mushrooms that apps had flagged as safe. Ohio poison centers handled over 260 mushroom-related calls in a ten-month period, with patients consistently reporting identification apps as their primary guide. Amazon was separately flooded in 2023 with AI-generated foraging books containing potentially deadly misinformation, some with no indication of AI authorship. Google’s AI overview once helpfully provided cooking instructions for amanita ocreata, a mushroom that destroys livers.

None of this, however, makes mushroom AI worthless. Platforms like iNaturalist, which combine AI suggestions with human expert verification and even DNA barcoding, represent a genuinely useful approach. The sensible advice from experienced foragers is to treat apps as one data point among many: use them to generate hypotheses, not conclusions, and never eat anything based solely on algorithmic confidence. Anyone who’s spent time in the field already knows this instinctively. The technology is a supplement to learning, not a shortcut past it.

Know your tool

The phone in your pocket can now pick out a scarlet tanager by its song, plan a bass trip, and (sometimes) identify a mushroom. It’s remarkably good at the first two, and getting better at the third, slowly, with appropriate disclaimers plastered all over it. The pattern worth noticing is that these tools work best when the problem is well-defined, the training data is deep, and a wrong answer costs you nothing worse than a missed bird. The further you drift from those conditions, the more important common sense becomes. Learn the skill, know the territory, and maybe don’t eat anything a computer told you was fine.