Suggested IDs could include similar species not yet recognized by the algorithm

Platform(s), such as mobile, website, API, other:
All platforms, particularly the mobile app.

URLs (aka web addresses) of any pages, if relevant:

Description of need:
Besides “expected nearby” and “visually similar” (but not expected nearby), suggested IDs could include an option “might also be” (or something like that) which would list other species not yet known to the algorithm but often confused with the suggested ones.

Feature request details:
If iNaturalist keeps track of suggested IDs that are manually corrected by experienced users (i.e. the algorithm suggests a particular species with high probability but the user ignores the suggestions and identifies the observation as something else), it is possible to learn that “observations of X are often misidentified by the algorithm as Y”. Then, whenever a user uploads an observation that looks like Y, in the “might also be” list of suggestions the user should also see X as a possible ID.

An example: the number of RG observations of Myosotis welwitschii is not yet enough to train the CV algorithm, therefore the list of suggested IDs never includes that species. The most common suggestions are Myosotis scorpioides, Myosotis laxa and Myosotis arvensis. I have already uploaded several observations of Myosotis welwitschii which I have manually identified as such, ignoring the IDs suggested by the algorithm. Based on that behaviour, the algorithm could have already learned that “what looks like Myosotis scorpioides, Myosotis laxa or Myosotis arvensis, might actually be Myosotis welwitschii”, so it could include Myosotis welwitschii in the list of suggested IDs even though that species is not yet included in the CV model.

This could reduce the number of wrong IDs by inexperienced users, encouraging them to explore the differences between those similar-looking species. This in turn could speed up the obtainance of enough RG observations of that species to train the algorithm.

your “might also be” suggestions would have to be suggested in the context of another taxon. would this other taxon just be the highest-ranking species suggestion from the computer vision results (either visually similar or expected nearby only)?

how are you proposing the “might also be” options be ranked among the computer vision suggestions? or are you proposing they be presented in a separate section from the computer vision results?

you might want to look at how the compare module works. that offers many ways to suggest possible species and may be what you’re looking for.

that’s not how the similar species algorithm works. it needs disagreements in actual identifications, not just a user not selecting a computer vision suggestion.


That would be the Similar Species option on a taxon page - to be offered in the Identify modal too ?

It could work in two steps.

  • Right now, when we click the first time to identify an observation, we get a list of species suggested by the algorithm (expected nearby or not, we choose). If we select one of them and then click again, we get a list of subtaxa (subspecies, if we first selected a species).
  • If this suggestion is implemented, the “might also be” option (a third option in addition to “expected nearby” and “not expected nearby”) could be unavailable the first time we click or it could be based on the first suggestion. If we select one species and click again, then the “might also be” option would list species similar to the one we had chosen before.

A separate list as happens already for “include species not expected nearby” vs “expected nearby”.

You mean the two arrow button? I use it often and it doesn’t seem to emulate the feature that I described. Either it only shows visually similar species already known by the algorithm or it ignores visual similarity altogether.

A few days ago I suggested some changes to the “similar species” tab. The two suggestions could be articulated.

too complicated.

what i’m saying is that if you’re going to ask for something that gives you similar species, this module seems like the place that could provide it. it offers different kinds of suggestions as shown below, and you could add similar species as another option.

1 Like

No, it’s not too complicated: as I explained, it’s already happening.

I agree: not in the standard ID field that we fill in when we upload the observation, but in this more sophisticated identification module. Considering that the “might also be” feature would be used by somewhat advanced users, it seems fine to make it available here in the “compare species” module and not in the upload module.