Does the computer vision suggest different taxa depending on the user?
I have recently seen a certain number of wrong IDs by one user that were suggested by the AI. These were observations of taxa that could have been “easy” to be identified and pretty common in that area.
The AI did not suggest the same wrong ID when I asked.
So I wonder if, at least sometimes, the AI takes into account who is asking to make suggestions for the ID.
No, it doesn’t take the user into account. It uses location, date, visual similarity, and the IDs that have been added to the observation.
There’s also a difference between using the inaturalist app/website and using seek. The image recognition built into seek is less capable than the main one (as far as I know).
The CV suggestions can change over time as more and more species are included in the database. E.g. it may have suggested something different two years ago than what the top suggestion is now.
It’s also possible that maybe the user had chosen to include suggestions not seen nearby.
Correct, Seek’s model is a few years old and doesn’t take location/date into account.
In the meawhile, this user is still identifying in a bizzare way plants, while I do not receive the same suggestions by the AI. Maybe he is really using seek.
On the observation page, look for a mention ’ This observation was created using: Seek’?
@SQFP Is that a phrase that is automatically tagged with such IDs or is it user-added? Where would I see that type of “mention”?
I have taken a look but there is nothing mentioning seek in these observations.
It’s automatically added by the app used to record the observation, I guess.
For example:
Sympetrum sanguineum from Linguizzetta on August 30, 2023 at 03:42 PM by ansgartk · iNaturalist shows “This observation was created using: Seek”
… whereas …
Tribe Carpocorini from île de Corse, Furiani, Corse, FR on August 29, 2023 at 11:19 AM by hpfioc. Nymphe V · iNaturalist shows " This observation was created using: iNaturalist iPhone App"
edit: fixed wrong 2nd link
For that first example, I finally found the “Seek” phrase in the lower right under “Copyright Info and More…” but I fail to find any analogous phrase on the observation page for that second one. Where are you seeing that phrase, “This observation was created using: iNaturalist iPhone App”?
I’m looking at these on the website on a Mac using Google Chrome.
damn, copied the wrong link. Sorry. Edited.
Also let me insert the obligatory “It’s not an AI” because it’s not – it’s CV. There’s a difference.
I sometimes have doubts that it could be definied intelligent.
Why and how and which ID’s that have been added to the observation are used? I was surprised to find out that a photo with a lichen and a mosses Computer Vison returned a moss if a moss-id was added and in a lichen if a lichen-id was added.
yes, this is a known way of ‘manipulating’ the suggestions. So if you photograph an insect on a flower, you can add an initial ID of either insects or flowering plants (or a finer ID within either category), and the CV suggestions will then be different
If it is a known way, did you know it by using of CV or is this feature documented somewhere?
https://forum.inaturalist.org/t/log-computer-vision-taxon-guesses/44224/1
A post was merged into an existing topic: Introducing the iNaturalist Geomodel!