I don’t know about that. Yeah, a lot of things I only ID as fungi or plants don’t go further…on the other hand, almost everytime I’ve IDed something as just birds, beetles, true bugs or lepidoptera (butterfly/moth), it will be IDed soon after, even if it was languishing as an unknown for months to years before my ID.
Just because there aren’t a lot of people going through “plants” or “fungi” now, doesn’t mean they might not get picked up like “birds” in the future, and staying at plants indefinitely is better than staying at unknown indefinitely.
True, but that isn’t immediately visible to the identifier unless they take the time to look at the comments
It could be so although these are such diverse categories that experts likely don’t look this broadly. For plants, one could be an expert in trees or legumes or wildflowers and not no much about others.
I agree with what another poster said about there being value in submissions versus nothing submitted at all. If we want to collect lots of data we need to make it accessible to the casual observer and understand that they might not take it as seriously as others.
After reading the entire thread I have noticed not much emphasis has been put on those vetting observations. Perhaps the real issue is not with those posting observations but a need to modify the process of observations getting to research grade, such as upping the number of identifications required?
Someone else alluded to this probably never happening but there could be a process where a user who consistently misidentifies things could gradually have less and less weight on whether or not something gets to RG.
As another user mentioned iNaturalist is not advertised as a highly intense means of cataloging carefully-collected identifications and rather a means of connecting with and documenting nature, I think that these things need to be accepted and the system of how casual observations get to research grade needs to be modified. In doing so the AI will get better and gradually over time there will be less and less issues.
OP I can understand your frustrations but your identification carries just as much weight as the observer’s, which prevents the observation from going to RG without further vetting. If people want to argue with you you are in no way obligated to explain yourself and respond.
I hope you’re posting. I think how quickly something goes from ‘dicots’ to a more specific taxon depends on the photo and the location. I feel blessed with a few people who seem to review a lot of my spiders and dicots and hover flies and manage to get them to a more specific taxon than I could. Many of them will even answer questions when I ask! But I don’t know that they review for the entire world and I think there are probably gaps in locations where they don’t have such generous people helping out.
I would guess less than 1% of my fungi have made it to anything more specific from fungi. I think that’s partly due to not having many people identifying fungi but I think it’s also due to needing more info than a photo can give. I have an awful lot of ‘dicots’ as well that might not be identifiable by anyone (due to time of year or needing more info than the photos provide).
But I post pretty much everything I take time to observe. I tell myself … someday, I might get real skilled in identifying sedges and all those photos stuck at ‘true sedges’ will get an ID. Or, someone might just come along that is really good at identifying sedges and find those old photos and identify them. (I also tell myself that … someday, I will see a fox in my urban backyard and I don’t have any reason to believe that it isn’t possible!)
But I also think I learn from all of these observations a little at a time. What I’m learning might not be the name of the species. But I’m noticing the progress of life throughout a season and how it differs in different sorts of habitat. And I have decided that I probably don’t really care what species of fleabane that is but I know it’s a fleabane and I couldn’t have identified it last year as fleabane.
And if only a smallish percentage of the things I take photos of get identified (at least in certain categories of life), then I figure if I just take a whole bunch of photos of different things, I up my chances at hitting one that might get a species level id.
But this is all for fun for me. I mean, I am having a ball (other than the bug bites which always seem to fester on me). I am that weird lady at the park that is prowling around the weeds then hunkering down to take a photo of something (they don’t know it, but it’s a lady beetle pupa!) when other folk are mostly looking at only the hero plants and birds (or their phones). Prowling for new blooming wild plants and unfamiliar insects and newly hatched dragonflies is the most welcome distraction for loads of other things I’d rather not think about. :-) I just think of every id I get as a new gift.
I want to thank all the specialists that take the time to correct my Id’s. I am one of those amateur naturalist enthusiasts that regularly explore the outdoors and when I see something totally amazing I snap a pic and use iNaturalist as a starting point. I enjoy and absolutely rely on what feed back and suggestions I receive on my “discoveries”. i love to learn new things and I thought I was being a citizen scientist by at least documenting the beautiful diversity in my area. Again, thank you to all those who take the time to ID! Thank you for sharing your knowledge!
People who ID set their filters at various levels.Plants, or dicots can and do get IDs. Even that VERY broad category is more visible than Unknown or Life.
There have been earlier discussions about weighting IDs but iNat has decided that we are all equal, no experts.
I very frequently use the AI for species that I know 100%. For a while I was doing it just because I was fascinated by how well the AI feature worked as was curious to see how often it was accurate. Even now I often use the AI out of curiosity of whether or not it will get a particular ID right.
I have to agree that really IS a problem when it happens, especially if both people never come back to change their identifications. Then it needs 5 correct identifications to get it over the 2/3 threshhold! I don’t see it often, but when it happens, it’s very annoying!
I absolutely NEVER type in a name in the app, even though I know (or think I know) exactly what I’m looking for. If the name I want isn’t in the ten suggestions, I just don’t put anything until I get home. I am not a millennial who can speed-type with my thumbs on a tiny screen! Even on the computer, I usually prefer to click on a name rather than type it in myself. The fact that someone accepts the computer vision ID doesn’t mean they didn’t know it independently.
Several of the keys on my laptop are broken so I have to use an external USB keyboard. When I’m being lazy it’s way easier to let the AI come up with the ID for me than to reach over and type on the external keyboard every time. Only time I don’t do that is when I know the subspecies or when the AI gets it wrong.
The problem with excluding things that are not nearby from suggestions is that means you won’t get suggestions that could indicate a new invasive.
E.g. consider the “murder hornets” that were recently discovered in the U.S. Their previous range was Asia. If someone in the U.S. had added an observation before that was made public, under the current system the CV could still suggest it. If it were excluded based on location, we would have to wait 6 months (when the CV is updated and after curators update the taxon’s range).
when I first started using this platform to record my observations, I was the first observer in my area for many, many, common organisms. The CV suggestions were not that useful. By far the majority of the suggestions appeared laughable to me.
As I continued to add observations and identifications; and as my identifications were confirmed (or not) by other users, the CV suggestions began to improve. This is what I expected and why this type of programming is described as ‘machine learning’. It appears to ‘learn’ - or change behaviour in response to input - in this case the input of my images and related confirmed identifications.
the improvement is not a change in the CV ability to identify any particular organism - it is a change in the CV’s ability to identify MY particular way of photographing that organism.
Take a moment and let that sink in.
The CV provides suggestions based on the probability that the uploaded observation image is more similar than not to other images which have been identified. The accuracy of the CV does not depend on any individual field mark to be visible -it does not memorize the number of petals or evaluate the length of a tail. These are things that people use to identify things. It depends on the number of similarities between the new image and the images already identified.
If the organism has a name (which they do not all do) and people agree on which organisms belong with which name (which is not always the case), then the only thing standing in the way of the CV improving is more identified images.
When identifying my own or others observations in my own locality, I check the CV to determine if the suggestions make sense. If things start looking wild, then I try to increase the number of observations of the organism concerned. Over time this has improved the accuracy of CV suggestions for commonly observed organisms in my area significantly.
The CV is a tool, like a field book or a magnifying glass. Like any tool, it can be abused or misused. To my mind it seems excessively harsh to advise that it should not be used at all, especially when it is so easy to improve its performance by adding more observations :)
If the suggestions listed Nearby first, followed by the visual match but not nearby - that would help. We still need to weigh up the suggestions before we decide.
These discussions are always very interesting, if only to perceive the variety of iNaturalist user profiles.
My small contribution to the debates: I am mainly an identifier, almost exclusively on European opilions .
Like everyone who speaks here, I regularly come across suggestions from observers, or obviously erroneous proposals from the CV.
Is it conceivable that these suggestions could be marked as systematically incorrect, in order to reduce the probability of their appearance in the automatically generated list?
I hope that the translator does not betray my thoughts too much!
The AI is only trained on species that have more than 100 RG observations (I think). Therefore, a low percentage of misidentifications won’t prevent the AI from improving.
i love to learn new things and I thought I was being a citizen scientist by at least documenting the beautiful diversity in my area. Again, thank you to all those who take the time to ID! Thank you for sharing your knowledge!