Below is @cmcheatle’s proposal, which I’ve copied and pasted. @cmcheatle let me know if you want me to change the subject.
Ideally the help page and/or curator guide should also be updated to specifically note that users who choose an obviously wrong ID for a human as a result of it being offered by the CV are not eligible for suspension to help document /reslove the constant flow of flags that come from this (as a curtator, I am perfectly fine updating how-to pieces of the document, but site policy stuff should be written by site employees)
I don’t think this is what @tiwane iwane was saying. I think it’s more an issue of situations like when a human is in the photo but not the subject of the photo. Why break the algorithm for those?
That’s not what was i was suggesting. What I am saying is there is a constant stream of flags caused by this, and that the policy that seems to have been adopted for dealing with them is that if a user picks an obviously false ID for a picture of a human from the list of computer vision suggestions, it is considered ‘testing the app’, or ‘taking the app suggestion’ and not an intentionally false ID, or cause for any action to be taken.
maybe i am misreading somewhere, but to me that looks like a totally different issue than the one being discussed here, which is why i was confused. I agree that it would be nice to have a clear policy laid out in how to deal with these, whether to flag, etc. But, while i don’t have much tolerance for intentionally bad data, i don’t think instant ban for first offense is really appropriate either. People are sometimes suspended for doing it repeatedly.
user adds picture of human
user runs computer vision on it and it suggests items other than human, and they pick one, either in joke, insult etc
result is bad data
users get upset about encountering them and flag the sightings
curators are apparently expected to dismiss the flags as not appropriate as users are ‘testing the app’ or ‘accepting cv suggestion’
@tiwane since you were looking at this thread can you add some clarity ? @cmcheatle seems to be saying that there’s some inat policy from the devs that people adding joke IDs should never have their IDs flagged and would never be suspended? I didn’t think that was the case. I thought it was another grey area where with one or two bad IDs we should ask them to stop but if it’s chronic or keeps happening they can be suspended.
From my perspective, if it’s a student making one or two joke IDs (that do not involve racism/sexism) on the own observations and it’s not a continuing behavior, I would just ID the observations as human, maybe make a comment, and move on. If the IDs are homophobic/racist/sexist/etc, or if they’re doing it a lot and adding those IDs to other users’ observations then I think suspension might be required, depending on the situation. So yes, I think it’s a gray area.
I have been going through these, and I really think that some people are being frivolous if not malicious. I saw several that appeared to be pictures of humans, identified as white button mushroom, dogbane family, and so forth. Is there a means of correcting obvious frivolous identifications?
Yep. I won’t say that people who choose “white button mushroom” when they surely know they are posting a pic of a human aren’t being “frivolous”, but I hesitate to ascribe malice when it was a CV suggestion…I think in those cases it’s more likely to be either silliness or carelessness.
From a bunch that I’ve seen, I’ve thought that even though the AI symbol is shown it could be resulting from someone typing in “white.” A lot of identifications look like they could have been typing in a race or nationality to get the list. My opinion is it doesn’t matter how it comes up–there was no good reason to choose anything but “human,” since it’s probably the most well-known species.
I agree, they definitely should not have picked anything but human, no matter what the CV brings up, whether out of the blue or in response to typing, but I was assuming a new user was picking the wrong thing by being too click-happy (or tap-happy, for the app?).
I hadn’t thought that they might be typing in a race, but that makes sense.
Another one I saw was for Gaylussacia frondosa (Dangleberry). That looks like typing in “gay” to me. Or it could have been the beginning of a name the subject has. Hard to tell.
I didn’t mean it was an insult, of course, just inappropriate in the context of an identification of an observation with a photo of a human being on iNaturalist. My main point was that I don’t think all these weird IDs are AI problems, but deliberately chosen in some cases.
Oh, definitely!
And even if they were pure CV suggestions with no user typing, the observer is still responsible for not choosing an incorrect suggestion just because, “hey, it’s not me, this is what came up!”
