I have seen this a lot while identifying fungi. People relying only on CVM to make an observation research-grade or taxon-level. Sometimes these identifications are right, but they can also be very wrong. How do you feel about this method of using CVM to identify? I think it is great for adding an observation, but identifications…?
1 Like
I definitely don’t think anyone should send something to RG based off CV. That being said, how do you know this is the case? e.g. many people select from the CV suggestions just because this is faster than typing, so the CV badge doesn’t always mean CV was actually used.
26 Likes
At best I consider it reliable to Genus if I can verify it to some degree and I have some knowledge of the subject. For example if it suggests 10 species in the same family then I will ID it to Family or Order or Class depending on various factors
4 Likes
You can’t really know if people are only using the CV, unless their ID is incorrect (such as picking something suggested by the CV, but which is not found in that part of the world).
I use the CV on occasions such as when I can’t remember how to spell something, or if I have a mental blank and know what something is but can’t remember the name. So if you see the symbol beside an ID I’ve added, you can’t assume that I used the CV without knowledge.
19 Likes
Yep. I’ll do this if I know the CV suggestion is correct.
If I’m uncertain I’ll back it off to genus or family level, but often that’s also an option in the CV menu, so even that likely gets marked as CV based.
13 Likes
Instead of setting ‘I can typing!’ and ‘spell scientific names’ against ‘used CV’ - there are better clues to use. New and enthusiastic and used CV - bit wary. Taxon specialist and CV is more efficient - not an issue. Use CV like a field guide, or a third opinion - it is still your name on the ID.
4 Likes
Obviously, one should not use only the CV to take the observation to RG. However, you can’t tell if that’s what happened because we also use the CV when we know what the organism is, just to make typing the name in easier.
13 Likes
Sometimes I think “I think this is an Asteracea” when it has just leaves, then hit “Suggestions” and see how many of the suggestions are Asteraceae. Sometimes I think “I should know what that is” but draw a blank, so I hit “Suggestions” and see if they refresh my memory. Sometimes I identify a higher taxon but say “CV says it’s Genericus but I don’t know that genus”.
9 Likes
I think, it depends on many things. For example the quality, quantity and reliability of the data basis. From my experience, iNats CMV can be a good start for an initial ID in many cases. It’s then up to you to start to search on available data for a more precise ID or just wait within iNat for correction/confirmation etc. With AI outside iNat, I would be careful, especially with comments that can be misunderstood as criticism. I don’t know what currently happens in the tech world, but I think a personal observation/ experience is something that should be still allowed to discuss in a Nature forum (if it’s a nature topic). For the background, just after my last post here, someone broke into my flickr account (I don’t have a direct iNat account) to get my e-mail adress and real name as well as going through my activity. The way I was informed by Flickr might suggest (not completely sure) that it came through the third party service (iNat/iNatforum).
Almost all of mine appear with the little badge at upload due to my process.
I have an extremely hard time identifying because so much of my knowledge is experiential. For plants, for example, the sunlight a plant grows in, the way a plant sits in the ground, the way the leaf array felt (sparse, abundant), what insects were on the plant, the smell of its leaves, how it moved or when the wind blew. Not being able to see these things in photos makes me get almost anxious and feel my lack of formal training utterly but sometimes I will feel brave and go identify something I feel confident about, like Tridax daisies. These are almost always already at Tridax procumbens but remain at Needs ID because few seem to care for them. I go through carefully, looking at the photos. Those that are T. procumbens, I click “Agree”. I have not relied on the CV, but rather on my vast experience of this elegant, little, bee-beloved bloom which I appreciate in all its forms. But you would have no way of discerning that.
3 Likes
Wait, what happened? I am not familiar with Flickr but are you saying someone from here, not staff, contacted you via your email? I am not sure how that fits with using CV for identification but if someone to whom you did not provide your contact information is contacting you and you think they could have come via here, I think I would reach out to iNat staff.
3 Likes
referring, apparently, to this topic
I concur; though it’s not clear to me when this happened (five days ago? three hours ago?)
Rather than flag these two posts as off-topic I’m going to message Anne to clarify what’s going on. Generally speaking, there’s nothing wrong with mentioning personal incidents of AI problems / security concerns but I don’t follow how that’s related to the topic here.
4 Likes
Not sure staff can do something here. Back to the topic. iNats CMV can suggest IDs based on fotos in an observation. It does not tell you that there are Champignons in your region and that you can eat them without problems. I think, it’s not that likely to confuse Agaricus species with Amanita species based on iNats CMV . What do you think, what are your experiences.
I think at no point should the CV or any other aspect of iNat, such as species information pages, advise on the safety of ingesting anything.
I am aware there are mushrooms here since we have a rainy season, but as I only have 21 Observations in the entire kingdom of fungi, I am still extremely unknowledgeable. (To be fair, I am distracted by the other species that come with the rains, like rain trees, rain frogs, rain lilies, dragonlets, etc.)
6 Likes
I am totally against using computer vision (whether the one built into iNat or something external like Google Lens) to identify observations made by others. I think it defeats the whole purpose of putting them before the community. “Computer vision said this was X twice” is not even close to the same thing as “two different people said this was X.”
That said, I agree with others here that the CV symbol appearing next to an identification doesn’t necessarily mean the person actually leaned on it.
We have no way of really knowing. Although I have strong suspicions in a few cases I’ve seen, like broadleaf trees identified as “eastern white pine” or an arborvitae hedge identified as “mosses.”
6 Likes
Although I’m much more comfortable with just visual clues than you are, I do find this true some times. If I could just feel those grass leaves! If I could stand in that site and assess what it is good habitat for! Sometimes I’m embarrassed that I really can’t identify a grass that I’m sure I would know with a few more clues.
7 Likes
When identifying, unless I’m very confident about the CV’s suggestions (mostly bird species I’m very familiar with, or there’s nothing else it seems it could be), I actively avoid using suggestions to make an observation research grade. It would be essentially an educated guess.
2 Likes
I generally don’t mind it if people stop at family or genus, but I think it’s a very bad habit to trust it for species ID, particularly when people do this for the observations of others and not just their own. My views on this have softened a bit, and I’ve come to realize that even incorrect initial IDs can be useful if they get an observation in front of the right eyes.
For example, morels and stinkhorns can look quite similar, but they are in different phyla. If someone can’t confidently tell the difference, the most “accurate” thing to do would be to leave their observation at fungi, where it will likely be lost among millions of others. If they make a guess based on the CV and ID their stinkhorn as Morchellaceae or Morchella, it will reach me, and I can provide the correct ID (because they are common lookalikes, so a specialist IDer inevitably learns both). If they withdraw their ID when corrected, the observation then gets where it needs to. Even if they don’t and it’s stuck at fungi, it now has two competing IDs that are easy for specialists to filter for.
But there is absolutely no reason to blindly trust it for species-level IDs, and even less so to do so for supporting IDs. People who do this in bulk can definitely harm the accuracy of the iNat data and consume a lot of IDer time (that’s not a judgement against the people, just an opinion on the effects of the behavior).
2 Likes
Also, I’d lightly disagree with some earlier comments that it’s impossible to tell when someone is doing this. Of course it can’t be known with 100% certainty, but if one specializes in difficult taxa, I think it can be quite obvious and is a reasonable inference. I would never assume this based on the presence or absence of the CV badge, but for semi-cryptic species without good keys, the only way to make a good ID would be with a lot of time spent reading and looking through a really large number of examples.
When you specialize in one of these taxa, you know exactly how much time and effort it takes to learn to ID them, and you generally come to recognize the small handful of people who are capable of making a good ID. Most of those people ID systematically, focusing on a difficult taxon or group of similar taxa in any given session. Many of them use CV instead of typing out the names just to save time, and I’ll probably start doing this myself soon. Yet if I click on someone’s profile and it shows that they always/often ID to species while rapidly jumping between difficult taxa with no clear pattern (jumping from morels to red Russulas to difficult insects), that tells me they either have a truly rare amount of knowledge and many years of experience, or they are an overenthusiastic (often new) identifier who thinks they can ID by CV and general appearance. I think one of these things is much more common than the other. Then there are the even more obvious cases. If someone IDs a pinecone, acorn, or piece of wood as a morel (I have seen all of these), I can safely infer that they were blindly following CV, as it would be hard to make that sort of mistake with a guidebook, key, or really anything else.
I’d be lying if I said that sort of thing isn’t sometimes frustrating, but it’s generally people who don’t know any better and have good intentions, and correcting the errors that result is just part of the game. Even without CV, these types of identifiers might be speed-IDing with outdated keys or based on misleading/broad common names (like “black morel” for Morchella angusticeps or “chicken of the woods” for Laetiporus sulphureus). Of course I do wish there was better onboarding with a bit more emphasis on what not to do, as it could perhaps reduce the scale of the problem.
4 Likes
Several times I’ve come across a thing dangling from an ant identified by CV as Ophiocordyceps sp.. I know that O. unilateralis has been split into several species named O. camponoti-talis-huius-et-alius, each of which attacks Camponotus talis-hic-et-alius, and they can’t be told apart without identifying the ant or getting the DNA analyzed or something, so I look for an associated observation of the ant, and if there isn’t one, ask if the ant is identified.
2 Likes