Photo gives surprisingly bad AI suggestion, when location is specified

Please fill out the following sections to the best of your ability, it will help us investigate bugs if we have this information at the outset. Screenshots are especially helpful, so please provide those if you can.

Platform (Android, iOS, Website):
Website

URLs (aka web addresses) of any relevant observations or pages:
https://www.inaturalist.org/observations/140547860

Screenshots of what you are seeing (instructions for taking a screenshot on computers and mobile devices: https://www.take-a-screenshot.org/):
(with location)
image
(without location)

Description of problem
On the Upload observation screen, if you upload the photo from that observation and set the location to Pawnee, Oklahoma, the AI’s only suggestion is “Great Egret”.

If the location is not set, the suggestions are much more appropriate, with it showing various types of grass spiders, etc.

Interestingly, once I created the observation, if I go to the “suggest an identification” on the new observation, it shows similar results to the “no location” case above.

I just found it surprising that the AI gave such poor results in this case, not suggesting a single spider species or even taxon when the location is set.

Welcome to the iNaturalist Forum @jesusfreke!

This is probably not a software bug, but I’ll let others weigh in too. When the Computer Vision (CV) suggestions are limited to a specific place, that reduces the number of species that the CV “knows about” to just the ones with existing iNat observations near that place. If none of the spider taxa currently in the CV model have existing iNat observations near Pawnee, OK (I haven’t checked that), then it won’t show you any spiders, and will show you whatever looks next-most-similar in the area. By removing the geographic restriction, the CV then had a lot more species to choose from.

So depending on where you are, and the observation density in that area, CV suggestions may be better with or without the geographic restriction.

Also note that the CV doesn’t know about every species, only those with enough existing verified iNat observations and photos to meet the threshold for inclusion in the current CV model. So always take its suggestions with a grain of salt, verify independently if you can, or just enter a higher-level ID (like “spiders”) if you can’t.

5 Likes

Also discussed here: https://forum.inaturalist.org/t/some-interesting-ai-suggestions-of-late/34847/7

@jdmore is there a way to have the CV at least go up to Family level of the first similar species it finds where it’s out of the local area and suggest that instead of going to completely remote taxa? An Egret suggestion for a spider (or see in the other thread for more) is entirely useless, but if it at least suggested Gnaphosoidea or Entelegynae it gives the user a useful starting point.

I don’t know enough about the inner workings of CV-user interface to answer that question. I don’t even remember the top taxonomic level it will suggest before it just says “not enough information.”

Why not start with an ID of spider, and then see what CV or Seen Nearby or Visually Similar can offer?

2 Likes