Best way to help CV correct suggestions that were mis-identified

I noticed a moth in my observation, the top suggestion AI gives was not correct, even the genus level “we are pretty sure” was wrong. I believe it was due to a couple of research grade ids that were incorrectly confirmed and that led to subsequent mis-identifications. If one spots a situation like this, what is the best way to help AI correct it?

I believe the only thing you can do is put your own ID on any observations you believe are misidentified, and hope the next computer vision model (there should be one in about a month) is better.

5 Likes

BTW, I have an issue with Senna pendula (distribution map) and Senna bicapsularis (distribution map). I put a bet on the hypothesis that there may be many cultivated Senna pendula in Asia misidentified as Senna bicapsularis. In most of my Senna bicapsularis identifications I add “else Senna pendula” in comment, but the CV training will not take this into account.

Could someone tell if leaves with 4+ pairs of leaflets with yellowish margin can still be Senna bicapsularis or is necessarily Senna pendula?

I will ask @alelimasenna in a private message (he may not be here in the forum).

(changed the title to “CV” instead of “AI” as “computer vision” is more accurate - it’s not an AI)

Yup, add waht you think are accurate IDs, provide evidence for your IDs, and maybe tag some other users to take a look and weigh in.

4 Likes

If you have a species for your moth - check the About tab on the taxon.
Included or still Pending?

Then click the Insect link in the next CV blog update.
https://www.inaturalist.org/blog/75633-a-new-computer-vision-model-v2-1-including-1-770-new-taxa
PS the conversation in the comments is interesting, to see where CV is making appreciated improvements.

If still pending … add more, good, photos till the CV includes the species. Very satisfying to see ‘my photo’ included in our learning curve.

4 Likes

Best way to get the CV better trained is the get more observations raided to accurate research grade.

The AI is trained on observations that have a certain number of research grade observations. If those are accurate and correct the CV gets trained well, if they’re not, then the CV is trained badly.

For machine learning it’s basically volume, accuracy, and variety.

More accurate RG observations = more accurate CV suggestions.

5 Likes

@tiwane Using “computer vision” may be more specific, but most sources consider computer vision a branch or subfield of AI. I’d say computer vision is an AI.

“CV” is an existing acronym with many more-popular meanings. Using it undefined in the title is a bit confusing.

Also, “correcting” user-generated content is generally a legally-tricky issue for websites. Ensuring iNaturalist remains covered by Section 230 exemptions may be important. Removing content is fine. Adding disclaimers or notes is fine. Admins editing content that then retains the appearance of being generated by a user is frowned upon. In addition to creating legal liability, it’s also generally unsettling for users to know that their content may be surreptitiously edited while remaining attributed to them.

1 Like

When a post is edited, an orange pencil symbol appears next to it

clicking that pencil generates a pop-up that very clearly shows what the edit was, and who made the edit (this is visible whether logged in or logged out).

I’m struggling to see how you could construe this as being the case when not only is the edit history openly available to any user as I noted above, but also

  1. When a moderator makes an edit like above, a notification is generated for the poster
  2. Tony explicitly stated in his comment above that he made the change

That a moderator made this edit, not the original user, is patently transparent here

4 Likes

Thank you for the info. I checked that species, it’s still pending. I wish I could encounter these moths more to add more photos, hopefully this summer. Interesting process.

1 Like

OK thanks. I’m not on the forum enough to know the correct lingo.

No worries! We used “AI” for a while but felt that “computer vision” is more accurate.

3 Likes

The orange pencil must be a moderator-only feature. Mine is grayed out.

I was unaware I could click it and see the edit history. Thank you for explaining that.

I’m still not a fan of the edit and don’t find it more accurate, but the transparency is there so that’s good.

2 Likes

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.