You’re in the woods. It’s October, the light is doing that golden thing, and there’s a mushroom at your feet that looks — you swear — exactly like the chanterelle in the photo. Out comes the phone. The app thinks for a second, then announces, with total confidence: chanterelle, 94% match.
Here’s the thing about that 94%. It’s the app’s confidence in its own guess, not a measure of whether you’ll be fine. And the guess can be spectacularly, dangerously wrong.
This isn’t hypothetical scolding. The consumer advocacy group Public Citizen recently reported that people have been hospitalized after eating wild mushrooms an AI told them were safe. And in a benchmark run this fall — 1,040 photos, 16 AI models, 55 species — even the best model got its first guess right only 65% of the time. When it missed, it missed the way that matters: a deadly webcap was identified as a chanterelle. That is, to be blunt, the exact mistake that kills foragers.
Why the apps are so confident and so wrong
A photo is the thinnest possible slice of a mushroom. The app sees cap, gills, stem — a flat image. It doesn’t see the spore print. It can’t smell the apricot sweetness that helps confirm a real chanterelle. It doesn’t know which tree the mushroom was growing under, or that the deadly lookalike’s telltale cup at the base of the stem is buried an inch under the duff where your photo can’t see it. It doesn’t know that in the button stage, a death cap and an edible paddy straw mushroom are essentially indistinguishable — a mix-up with a body count.
And the apps have a structural problem: they’re trained on photos, and photos lie. Lighting, angle, age, weather — a rain-soaked mushroom is a different visual object than the same species on a dry day. The training data skews toward common, photogenic species from well-photographed regions. Your weird little brown mushroom from a ravine trail? The app is extrapolating.
So are the apps useless?
No — and that’s what makes them dangerous. As a starting point they’re genuinely handy: “this looks like it might be in the Russula genus, let me go read about Russulas.” The mycologist Dr. Gordon Walker has said iNaturalist is the only AI-powered ID service he’d even consider using, because it’s indexed on local species and curated by actual experts. Note the framing: the gold standard of the category is still a tool, not a verdict.
What to do instead
The unsexy, correct answer:
- Use the app as a suggestion box, not a judge. Let it narrow the field, then verify against a real field guide for your region.
- Learn the spore print. It’s free, it’s weirdly satisfying, and it eliminates entire genera at a stroke.
- Find your local mycological society. Ours is the Vancouver Mycological Society, and societies like it run forays where you can put your finds in front of people who have been staring at mushrooms for decades. There is no app update for experience.
- When in doubt, throw it out. Every experienced forager has a compost pile full of “probably fine.” That’s not failure; that’s the sport.
The rule we live by on the show: no app, no photo, no confident stranger on the internet gets a vote on what goes in your mouth. Dinner clearance comes from your own verified identification — or you admire it, photograph it, and leave it for the slugs.
If this has you curious about which mushrooms are actually worth learning first, Episode 3 — “Know Your Edibles” — is exactly that conversation.