Why so many implausible identifications?

Why so many implausible identifications?
It happens to notice a relative high number of implausible identifications. In other words, observations of plants species that are more or less common and that are usually correctly identified by the AI, have received by users a genus or species IDs that are almost impossible to find in that given area. I would also say that these “exotic” taxa should be likely unknown by most non-experts of floras of distant countries. Finally, they are not even listed among the options proposed by the AI as a possible IDs.
I wonder if the AI works in different ways depending on the site/app or if it takes into account some “hidden” factors.

NB: if this topic has been already treated just delete the thread.

4 Likes

Were the ID’s suggested by the Computer Vision, or did humans make them without computer assistance?

I know I’ve made wildly wrong identifications because (1) I made a mistake about identification or (2) I clicked on the wrong name in the list and didn’t notice.

4 Likes

Seek does have a different CV model (older, simpler) than the apps/website I believe.

4 Likes

For which plant species have you noticed these identifications?

Just curious.

1 Like

You might want to read through this topic: https://forum.inaturalist.org/t/possible-increase-in-cv-errors-around-organism-range-location/34411

2 Likes

I often have to go through some frustrating AI fails, that users choose because it is the top choice. I know random Geo caterpillars are called the European “Black Looper”, and naked pupae are often called a specific European Cossid moth. I have to keep eliminating them to try to get the AI to stop suggesting that for an American moth.

1 Like

Are you noticing this with any species in particular being IDed more often, or is it pretty random?

2 Likes

I have sent you a private message.

I have not made a in depth study of the phenomenon but just noticed it…

Sometimes iNat, the camera GPS, or user error, causes species to get tagged with a wildly incorrect location during the upload process, and the user doesn’t notice it. This has happened to me a few times where I could trace it to the photo selection process on the phone version of the app. When you have taken a long trip in areas with poor cellphone connectivity you can pile up hundreds of observations. Then when you have connectivity you have to scroll through those hundreds of photos on the app and occasionally you accidentally select a photo unrelated to the observation. That photos lat/long data can then “contaminate” the observation you are trying to make, even if you deselect it. I don’t know if this phenomenon is related to yours, but it might be worth a check with the observer.

4 Likes

Have they used the CV or not? You can tell if they have by a little icon in the top right of their ID (on the website)

1 Like

Sometimes it may be accidentally clicking the second on the list. Has happened to me a few times. Multiple species so I choose Life - but the accidental second on the list is someone’s hand holding a ‘something’. I check that iNat is doing what I ask it to each time (slow and glitchy internet) - so I notice that ‘I have added’ an ID for a species I don’t know from cheese. People are rushed or tired or distracted?
Life brings up a very strange list. Then the homonyms, Erica, no not the spider …

Yes, those observationns I have succeeded to find (they are now IDed at a higher rank taxon level) have the shield icon shown in the implausible IDs.

1 Like

It is disconcerting to get a Seen Nearby suggestion from one, single, tentative ID.

3 Likes

I would think that many of these erroneous identifications are likely misclicks, especially since some species names are so similar. Another issue is that some taxon look very similar to a wildly different taxa genetically. I would think many Lotus observations are identified as Nelumbo and vice versa. Then there are currently multiple speed ID’ers on Inat, who just click “Agree” with everything solely to climb the leaderboard. This could create false CV suggestions. Finally, cultivated plants may get accidentally marked as “Not Cultivated” and cause a disagreement there. There are other reasons, but U would think these are the most common.
Hopes this helps.

By the way, anyone can add a taxa to a locale, regardless of Research Grade or not. Theoretically, someone could say that Spix’s Macaws are native in NYC, and it could show up as native on checklists.

We do see cases of young students looking for the weirdest possible names for each other’s photos. Kind of amusing, though also annoying.

1 Like

I’ve seen a lot of humans identified as monkeys and fictitious identifications like the famous “Gerald”.

2 Likes

With your plant example, it could be as simple as that rarer species are more exciting, so maybe there is some inclination to identify them as those rarer, more exciting species than some seemingly mundane, common species?

1 Like

People may have seen a picture of some species which has some characteristic vaguely similar to what they are observing, so they identify it as that. An example would be the numerous tree cankers and burls identified as chaga, because chaga is the one people have seen pictures of. It’s the classic bias toward what you know about and against what you don’t know about.

2 Likes