This is very concerning to me. Like extremely. Maybe this is why I’ve been seeing an increase in Chironomus crassicaudatus miss identifications in the past week. Without suggesting the higher taxon like Chironomus, or Chironomus Group. There will absolutely be an increase in miss identifications and I struggle to see why this was changed like this.
Honestly as an IDer, I wish the CV didn’t have to feel like something to battle and fight against. But certain features and the way it’s implemented makes it feel like that. As an IDer I should not have to be concerned about how IDing a species can make the CV much worse by creating more miss identifications that I have to work through. This feature just enhances that feeling and makes me even more hesitant to identify certain IDable species.
Example, if there is a worldwide genus with very hard to Identify species. But one species is unique and IDable found in only a single part of the world. IDing that species and thus training the CV will stop the genus from being recommended everywhere. The genus will just be unlearned, those other species not IDable in the genus will likely be miss identified as other things. This has already happened a few times within Chironomidae. The genus Chironomus, the most widely seen Chironomidae genus has been mostly unlearned because of this.
I will probably have to reconsider IDing certain species unless I can guarantee a few species within that genus can also be IDed. Which is actually pretty saddening and not what one would expect on a citizen science site. I can only hope that the future of the CV on the site will be improved.
So much could be improved if the CV system was expanded so it does not unlearn higher taxa, like genera, sub tribes, tribes, etc.
Living in a part of the world that has fewer observations than many others, I dislike the idea of making CV front-and-center of the process of adding an ID, since over here it works poorly at best.
Besides my pathological dislike of AI, I am concerned that this will result in an even larger increase in buddies supporting each-other’s CV-generated suggestions to produce even more wrongly-identified, ‘research grade’ observations. This is already a problem, particularly with school groups.
I tried making a mock-up of an identify-menu that I’d prefer. I think, just moving the taxon-search bar to the main page, instead of hiding it behind the loupe icon would mostly solve this issue:
Also, instead of the loupe icon, which would now be superfluous, I’ve added a question mark (like in the new audio-recording menu) to the top right corner. Perhaps this could open a popup with tips on how to identify. Inspired by the audio-recording help, it could maybe say something like this (very rough draft):
ADDING IDENTIFICATIONS
…
Add your identification:
– use the taxon search bar to search for a taxon and add it as your ID
– you can make IDs as broad or specific as you like (e.g. “plant”, or “insect”, or
– “Columba livia var. domestica”)
…
Top AI suggestions:
– these are the taxa our Computer Vision thinks are the closest match
– don’t blindly trust the AI. Try to verify an AI suggestion before you agree with it
…
Only ID as far as you are confident:
– It is okay to make mistakes, but if you are unsure whether your gut instinct or
– the AI suggestion is correct, but you are certain that the organism belongs to
– the same group, rather add the broad ID instead of the more specific ID
Personally, I don’t really share most of the concerns you listed. I don’t think the new layout would cause that much of a problem if implemented (though, admittedly, ladybeetles of the Coccinellini tribe are mostly far easier to ID than Chironomidae and there are probably more identifiers).
My concern was mainly that we would likely need more identifiers than before to correct mistakes and double check RG observations. But I feel like for most common taxa there are already identifiers who also go through already RG observations and add their IDs.
And for hard-to-ID taxa, the CV results would remain pretty much the same as before as @paul_dennehy showed: https://forum.inaturalist.org/t/the-role-of-cv-in-the-new-app/55349/10
I also don’t agree that the CV is something identifiers battle against. I think if anything we have to “battle” against observers being overconfident or overtrusting, which can be solved by better tutorials and education (see the help-box suggestion in my previous post), but is also part of the “job” of being an IDer, IMO. If the CV didn’t exist, we’d likely have a hard time even finding most observations, so I think it is definitely a net positive by a long shot.
Do those species have an atlas? I’m not sure exactly how this works, but if the distribution/range of the species are defined, surely the CV wouldn’t suggest exclusively Australian species on European observations? I thought that the CV had a “nearby” filter, in any case.
I sympathize with this feeling (as I’ve commented on other occasions (e.g. 1, 2), some of the peculiarities in the way the CV is trained are a big problem for bees, which present similar identification challenges as chironomids), and it is discouraging to realize that even if I educate one observer about why they should be cautious when using the CV suggestions, tomorrow someone else will come along and make the same mistake. And the CV doesn’t seem to improve.
But…this is probably getting rather off topic. It is an issue now; the current display of the new app may make it worse, but the underlying problem is the training of the CV along with users not being critical about CV suggestions. I have some ideas about concrete changes that might improve the CV training, but I find I need time to formulate my thoughts so I haven’t gotten around to starting a thread on the topic. It has been coming up quite a bit recently, so I think it would make sense to have a dedicated discussion rather than digressions in existing threads.
etc. Since plants are restricted to kingdom - we could have 4 icons available - and - all - at kingdom. Plant - animal - fungi - etc. Should do it. iNat says it works for plants.
What is this search filter? Filtering for app? Is this only counting active IDs? Well I suppose it’s not from Next then which is honestly good. Also yes C. crassicaudatus males are IDable hence the two IDs from me.
draft mode already effectively exists in the app already if you turn off auto-syncing, right? i’m not sure how that all works in Next, but i would have assumed that that still exists.
from what i saw, it takes significantly longer to add it as an observation note (that’s true for the old app too). Also, it ends up as an ‘unknown’ then which people seem bothered by per that other thread. I guess i could ALSO add iconic taxa but that adds more time. I know maybe the time addition seems trivial but i’m a generalist who moves through areas fast and for my decision making process, no option for the placeholder effectively means i will just not observe things nearly as often in an area with no cell service. It’s unpredictable what the algorithm can or can’t identify in areas with no cell service so it just adds a bunch of time for things that don’t offer much return. I realize i’m not a typical user and just one of millions of people using the site, but to me it’s off-putting enough as to change my behavior and seems like it might be to others, so it seems worth mentioning.
One thing that would make it easier would be a more streamlined review procedure before uploading. I know i mentioned it elsewhere but being able to swipe left and right in the observations in the app before uploading would help with that a lot. Also perhaps a color code where unknown observations that haven’t been uploaded yet show as red, ones with coarse ID show yellow, and ones with species level ID show green. But again, i recognize i’m a ‘weird’ user and maybe no one else would use these features and so it may not be worth dev time to think about them. To me, speed and efficiency are key. There is just nothing else on a smartphone that works nearly as well as inat for gathering ecological information. Things like survey123, field maps, etc just don’t work as well in my experience. But… to be honest, if it were just me i’d prefer having the app from 2012 or something, back when it had less to it and was just really fast to add things while walking by.
This is huge too. people don’t like it when this happens, but with the app very difficult to review things and with no ‘draft’ type mode on the website, it’s kind of hard to avoid
Yes, the choices when creating an observation are “Save”—which lists the observation in the app but it’s not uploaded—and “Upload Now.” If you’ve not added an ID yet, both of these choices appear the same color (white on our phone). Once you add an ID, the “Upload Now” button turns green.
yeah i don’t find the app currently an effective ‘draft mode’ for me, it’s too clumsly to review stuff (both the old and new app). I think it could be that way if it had some updates to it, but i don’t know if that’s feasible.
Sorry for the triple post here, i’m having trouble keeping up with the multiple threads and repiles and scrolling back up for multiple responses in one post.
I can not imagine a world where if you remove the top suggester which has always been more broad when the CV is confident enough. That an increase of miss identifications will not follow. Here it is family Curculinonidae.
For this observation. I can choose this higher taxa ID which is normally more right then wrong when it is suggested. Rather than choosing a species ID, or typing something like beetle. Without a higher taxa ID, I do not know how I could even use the CV when uploading. I like choosing the higher taxa because I don’t know enough about butterflies, and other stuff to confidently choose species. I will only usually only choose species if the CV is so confident it’s not even showing other results.
For your example I have two options. Either I trust the CV with a species ID, which could be wrong and easily turned RG by a single person. Or I manually type in something I’m comfortable IDing. I do not know these, so what should I do, type in flies, or the genus name of the species that the CV is suggesting?
Even if you do the later and choose the genus name. It will create so much more work for uploading. If I need to upload 100 observations, and I have to manually type in a genus or family name for each. That could easily balloon the time for uploading like 10-20 minutes longer. What if I need to upload 300? The amount of extra time quickly adds up.
This is some of the reasons why I think a broader suggestion is critical.
I looked into this search filter and there are only 29k observation in total with the appID filter. For Chironomids there are 39. 38 of them uploaded before September, some a few months ago. I’m assuming they were uploaded in the Beta? Most had broad initial IDs to be fair, but there are only 6 observers. So not really enough to conclude really much of anything.
I agree with you that the top suggestion should be kept at genus at most (as I’ve said in my post, I always thought this was a good choice and set iNat apart from other identification apps).
My mock-up was mainly intended to show a better placement for the “Adding your own ID” feature, so I just moved all the CV suggestions lower and out of the way
I feel you so much on this. Far too much of my time is spent fighting an ultimately losing battle against CV misidentifications.
With how often the CV gets things wrong or too specific, any feature that directs naive users to just select a CV species ID instead of a more appropriate broader ID or entering an ID not on the CV list is a massive quality-of-life downgrade for identifiers.
This is potentially disastrous. Take the genus Lyssomanes for example. There are about 100 species of Lyssomanes, but only 2 are included in the CV. This isn’t because the other species are rare, it’s because they are virtually impossible to ID by habitus photos alone. 90% of Lyssomanes observations should be IDed as “Lyssomanes” and no further. Under this system, they will all be IDed as one of the two species in the CV.
