I think most people would tend to simply dismiss a pop-up warning to not rely solely on the CV, the same way we don’t read data privacy notices on websites or “contents may be hot” warnings on food items.
The problem is not simply that people rely uncritically on the CV. They also reflexively agree with IDs offered by other users. Or they upload observations with no ID at all, not because they have no idea what they saw, but because they don’t know what to enter as an ID (e.g. they assume “bird” or “plant” is so obvious as to be nonsensical).
So I think there’s a larger underlying issue and that is getting people used to thinking about how do I figure out what something is. iNat’s interface doesn’t really encourage this – IDs are suggested by the CV or by other users as if by magic, without making visible the process of arriving at an ID. It also is a fairly passive form of interaction that doesn’t empower users to recognize that maybe they know more about how to classify things than they realized.
What I was thinking of was something more like a set of guided questions that users would answer when uploading an observation before getting the CV suggestions. (Experienced users would have the option to skip this and new users would also be able to skip it by typing in an ID themselves.) For example, they would be first presented with the iconic taxa and asked to select one or choose “I don’t know”. For each of the iconic taxa there would be some relevant additional questions (e.g. “Does it have wings?” “How many legs does it have?” or “Is it a flowering plant”?) that would help narrow the ID down to the rank of order or thereabouts. Only after this would they be told “Our computer algorithm has analyzed the photos and based on your answers and the reference images in our database, we think it could be one of the following.” In addition to the suggestions, there would also be an option “Don’t see any good matches? Choose [order] or enter a taxon of your own”. The idea would be to encourage users to take a more active role in deciding what ID to enter, as well as making it clear that the CV is not all-knowing and they are not required to use the suggestions.
This is also a really good idea. Something like a guided tutorial onboarding for how to use the CV for first-time users, walking them through the identification process.
For example, iNat might use eastern white pine tree as the onboarding example:
“Let’s say the computer vision model has suggested eastern white pine. Now, what should you do next?”
“Can you only confirm that this is a plant? → select plant kingdom”
“Can you only confirm that this is a conifer? → select conifers”
“Can you only confirm that this is a pine?” → select pine genus
“Can you confirm that this is an eastern white pine?” → select Pinus strobus"
“If you want to know how to tell eastern white pine from other pines in your area, please consult field guides or ask someone knowledgeable. In the mean time, make sure leave the ID only at a level you feel comfortable identifying yourself.”
The difficulty is that to be automatic, it would need to go through every taxonomic level, and some taxonomic levels are represented in specialist terminology.
Even with the example of pine, the question series would also include the levels “Pinales”, “Pinaceae”, and “Quinquefoliae” which the average person doesn’t know, even if they might know what an Eastern White Pine is.
It would be even more difficult with insects, mushrooms, etc. “Pterygota”: is this a “winged or once-winged insect?” Well, it doesn’t have wings, and probably never had wings, so how is it “once-winged”? Experts know what this means, but the average person doesn’t.
I think we could at least do the iconic taxa though, which would be immeasurably better.
And we could include all the intermediate taxa in the suggested list. Experienced users would use this as well. I often have to type in such things as “Pinaceae” because the CV doesn’t offer them, preferring to offer specific species of evergreen that a user is presumably expected to choose from.
CV has a lot of qualities: it is fast, always there, lack any bias, offers left field options to make me think and works fine for adult birds.
It is definitely not a guide. It ignores taxa that do not have enough RG observations to be included. If CV was validated against related taxa and genera, it would not send observers up the garden path.
There are taxa level suggestions with no location provided, which should not happen either.
When I started, I looked at other photos of suggested taxa and the rest of genus, sometime following up the “Looks similar” tab. I was more confident in my often incorrect IDs than I am now using keys and other resources. It took me months of googling the terms in helpful comments and reading up on subjects to realise where am I on the learning curve.
The iNat guidelines set clear ethical standards for participation and were not intended as an identification guide. There is no mention of keys and nor any general direction how and where to find resources. Combine this with the language that implicates that taxa level research grade is the end game, the pressure is on to agree to CV suggestions.
With no direction and no resources learning is by trial and error and lots of them, even for people who are interested to learn how to identify, Telling people to be more cautious, than the process that produces the suggestions, does not work. It is like telling people not to think about the elephant.
Why not make CV more restrained, to give genus level or higher suggestions to hard to identify taxa?
While I get your point here, I don’t think CV is completely unbiased. It will naturally pick up biases because the data it is trained on is based on humans with biases (if I understood how CV works correctly, at least). Unlike a human it’s consistent with the biases it has, though.
The above ten posts were originally slotted after this post in this thread, then moved to another and now moved back. Unfortunately Discourse didn’t allow us to put the posts back in their original positions.
Instead of reward mechanisms and facing Goodhart law, I believe the best solution is to allow people to source and have some global notes page for each taxon. Maybe it would be so haphazard at the start, but knowing little tricks for species goes beyond the Wikipedia page linked to taxa or even entire photos.
For example, knowing that the black drongo has a white rictal spot and is the only possible such drongo in that location will greatly help any newbie coming to the platform that a taxa photo page may not always convey directly. Similarly, for example, knowing neck plumes exist only for little egrets in India and throat markings in lesser-golden woodpeckers… there are a lot of such Knowledge clues that an average new iNat user misses when IDing, and I am sure any expert identifier who winces with mis-IDs wants to say something every time. I feel like it is passing on a better torch to recruit new identifiers to feel they are learning things over blindly trusting CV which does not explain its decision (as of now), especially for genus where there is often little to no information online to precisely know what to look for. I have seen one or two such photos showing such markers on taxa level (i dont remember but it is for some moth genus) and I loved people curating them.
I have seen ID’ers wanting to know how to distinguish when they make mis-ID and even though one can chalk it as a google search or species book would be a better solution, I feel there is still a better chance to concisely say some notes in a taxa level here on iNat too.
i feel the above can be two-way process between better CV maybe? and then users learning and jotting such clues to notes then.
As spring ramps up here and I’m trying to finish all my winter projects/help organize my local CNC/deliver specimens to the herbarium/collect more bryophytes/start a flora survey with colleagues/rip out and replace my kitchen/whatever this week throws at me, I sort of wish making IDs was NOT so inherently rewarding. There are so many to be made and so much else to be done in life (hey, I forgot getting out and making my own observations) that trying to keep up with the IDs I can do in my region is, well, overwhelming. Please don’t transform identifying into something even more addictive.
A simple free text community editable notes on taxonomy nodes would be very useful indeed.
As you proposed, these could provide identification clues.
It would be a good place for references for public domain keys and other resources and also what to record in an observation to make it identifiable.
I imagine this would be hard to quantify, but I’d be curious whether the identifier profiles posted periodically on the iNat blog have a positive effect on the overall IDing activity for the identifiers’ taxa of interest – i.e., do they increase awareness or waken interest in those taxa, encourage other users to gain expertise or contribute their existing knowledge, or help strengthen existing IDer communities?
Yep, that would definitely happen. I spent a year believing that there were Malabar whistling thrushes in my backyard because of a merlin (the bird id app) id. I never read the small pop up that read ’ merlin knows 5% of the likely birds in your area," and “merlin bird ids are not that accurate, so please only upload checklists with birds that you confirmed are really there”. Idk why because reading those pop ups are really helpful, if not easy, yet no one reads them.
The hard thing about that is most of the time, a casual person without too much knowledge on a species would have no idea that photographing that part of (insert species here) would be that important for identifying. Or maybe the creature ran away before you could photograph it well enough to show that important-for-identification part (eg a blue tiger butterfly’s lower wing,). This makes it harder for newbies like me to get good photos or identifications of (insert species here).