Platform(s), such as mobile, website, API, other: All (Suggestions)
URLs (aka web addresses) of any pages, if relevant:
Description of need: Most images taken in nature show multiple taxa, so iNaturalist computer vision faces difficulty in identifying the intended subject. This is a problem in any diverse habitat but I’ll focus on the problem I’ve noticed in terrestrial Observations. When plants are clearly evident and identifiable in an image, iNat’s model will sometimes Suggest only symbionts, e.g. rusts, galls, miners, etc. Naive users will not understand what they are agreeing to, and even expert users cannot force the model to identify the plant as opposed to the symbiont. Example of the problem below. An inexperienced user uploaded a picture of Glechoma, an extremely common and visually distinctive plant in our area. All of the Suggestions are for rusts and mildews.
[https://www.inaturalist.org/observations/245115697](https://Example Observation - Suggestions List Ignores the Intended Subject)
Feature request details:
In the Suggestions list, major taxon icons should appear, so that for instance any Plant suggestions have the leaf symbol while the mildew suggestions have the mushroom icon.
When there are competing major taxon Suggestions, add major taxon headers that divide the Suggestions list.
Plant Icon header. Plant Suggestions under it.
Fungus Icon header. Fungus Suggestions under it.
Don’t Know / Other header. Remaining Suggestions under that.
I approved this request for publication, but in this example reason you saw only rusts and mildews as suggestions is because the website filters by the observation’s iconic taxon (if present - in this case Kingdom Fungi). I’m not sure why the observer chose a fungus, but if you load the photo into the uploader and give it the exact same location, the top suggestion is Gleochoma:
I like this. It feels like an elegant and useful fix, and having a little iconic taxon symbol next to the name would I imagine at least help somewhat in those cases where an inexperienced observer wants to ID a plant but the first CV suggestion is a leafmine or mildew with no common name.
I also think a way of being able to initially limit the CV to an iconic taxon of the uploader’s choice (or have headers, as you suggested) would be very helpful (I know you can do this in the suggestions tab of the ID interface, but it would be great to do it while uploading on the website). If a plant is absolutely covered in leafmines or galls, it’s my experience that the CV will often not give a suggestion at all for the plant, which is unfortunate, since noting the host plant is often essential for IDing them.
However I would say, in the example you gave, there is white discoloration on the leaves, and the powdery mildew suggested (Neoerysiphe galeopsidis) is found on Glechoma hederacea, so even if that discoloration isn’t powdery mildew (not my wheelhouse), in this case I’d probably ask the observer if they were interested in IDing the discoloration or the plant, rather than assume an ID counter to what they suggested. If there’s, e.g., a leaf with no visible leaf mine, then IDing the plant (with a comment noting that you don’t see a leaf mine) seems appropriate.
In this and other cases, I know the users, and know they are not aware of or interested in plant symbionts (they are students just beginning to learn their plants). I don’t think there was any user involvement, the fungi or leaf miners were automatically chosen by the CV with no plant taxa suggested.
Thanks @jonsense for linking to those other threads, they’re exactly relevant. What we are seeing is the difference between AI and human intelligence. With millions of low-information users, the only way to address this is with a workable human intervention to clarify what’s intended. I’ve seen the suggestion that people ID a major taxon first, and then look through the suggestions that will be provided within it. I will try to work that into my instructions for new users (in my case, students). It’s a difficult situation, though. Some interactive logic would be great where first the user selects the major taxon, and then the CV suggests matches within that, or suggests alternate major taxa if the user is likely to be wrong.
You could show your plant students how CV works.
No indication of what we are looking at (most obs photos show a variety of biodiversity - beetle or flower?) or where it was seen
versus
a daisy in Cape Town (add your own good local example)
CV is pretty good, and adding about 1K new sp each month - if you start it off with what and where.
CV suggests, but it is the user who chooses.
I like this not only for the AI requests, but also for the searcher, there are a lot of organisms where the genus is the same in a plant and in an animal, examples like Viola, Ammophila, Knowltonia, Amaryllis, Diplotaxis etc. In this examples sometimes the person knows one example and not the other (botanist know the plant name but not the insect for example) and it gets missID. Usually is a very minor inconvenience, but sometimes can get accentuated quickly with “mindless” agreeing.
