Bug or problem?

https://www.inaturalist.org/observations/187581176
as i was going in the unknown section of INat, i was abt to suggest something for this observation, in the second suggestion the AI makes is “Dionaea muscipula” which is the venus fly trap, and it’s written in the front of it “expected in the proximities”
How tho? Venus fly traps are not native to Brazil, and Inat shows no records of them in Brazil, and specially in that area that is cerrado, the opposite of a bog or something that you would find a Venus Fly Trap. I’m very confused by this information, truly

see https://forum.inaturalist.org/t/introducing-the-inaturalist-geomodel/45407.

not a bug exactly.

1 Like

so it’s a problem? because venus fly traps should not be expected in that region

you know this, and i know this, but the geomodel AI couldn’t figure this out based on its inputs (observations and elevation). its predictions are apparently reasonably good for some species, but obviously they aren’t always great.

The goemodel includes cultivated / captive observations in its training dataset, so if there are enough cultivated venus fly traps in an area, the geomodel will include it as expected nearby. Same for lions in zoos or other houseplants.

The feeling about this are mixed… I personally like it’s a good policy since so many potted plants get uploaded to iNat, and if the geomodel excludes cultivated plants for an area, it would misidentify them. But other people are upset that the geomodel isn’t only for wild plants.

4 Likes

for reference, the geomodel for Venus-flytrap – it’s bizarrely worldwide in its predictions

4 Likes

It’s especially funny that it also predicts it as Johnson atoll in the middle of the pacific which has no permanent settlements and almost entirely wild observations, principally of birds.

1 Like

oh i see it

OH MY GOD LOL, they are everywhere, this is specially funny cuz since last week i’ve been trying to germinate them as a house plant, but still so crazy

Version 2.8 of the model was released yesterday and it’s looking a lot cleaner than the previous geomodel version. I don’t know the details but some of the thresholds have been tweaked.

Version 2.7:

Version 2.8:

I do want to emphasize that this is a work in progress and we’re constantly trying to make it better. And what’s better for some taxa might be not as good for others.

2 Likes

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.