CV is just a tool, like field guides or online photo references (including other iNat records), that you should use along with what taxa are reasonable to expect in the location where you took the pic. I don’t just blindly use what the CV suggests … I see if these other sources of information align with the CV ID before I accept it.
Of course, your local or not-so-local human with knowledge of the taxon is also an important resource … I won’t say tool, because that would be insulting.
Right. The distinction between artificial intelligence and machine learning is a real one. You know how you can search for something on your phone, and then the banner ads on your synced desktop suddenly have to do with what you searched on your phone? That’s machine learning.
Now, if it could understand that once you have purchased the product, you are no longer in the market for that product – that, I might call artificial intelligence.
The best course of action, I think, is to go with whichever one is correct in the case at hand. :-)
I analyse situation when humans propose ids, but will never rely on cv.
Both are continually learning. :)
I always defer to real people when it comes to taxa I have little experience with. Typically I will take the CV suggestion even if I know exactly what it is, just to save from typing it out. I know how to ID a snowy owl for example, but it’s way easier to just click “snowy owl” when it comes up as the first suggestion than to type it out every time.
When it comes to things that I’m entirely unfamiliar with, I’ll look at the suggestions and then work my way up to a classification I’m sure of. If I’ve found an unfamiliar wasp, but I suspect it’s in the family ichneumonidae and most/all of the species suggested by the CV are as well, then I’ll ID as ichneumonidae instead of trying to guess between which of the CV’s suggestions is correct. It can also a good jumping off point if it only IDs to a genus or family, if I have a book that has descriptions of species in said genus/family. If I find a syrphid fly I don’t recognize, but the CV gives me a subfamily as a suggestion, I can check out that subfamily’s section in my syrphid book.
tl;dr, it’s a useful tool and can speed up posting and IDing, but I won’t take its word as law.
I think, the CV is often a good start to get observations sorted into the right family/ genus. Especially with families that I’m not familiar with. Last year, I took part in an Mycoblitz and I used for the CV for some eye-catching fungi. The CV was about 50% right. On the other hand, I do not use the CV for mosses in my region because I know there is a skilled Identifier that asked not to use the CV. As long as this identifier helps me to identify mosses, I’m happy with that. When I use the CV on an animal/plant/fungi I’m absolutely not familiar with, I’m aware that it could be wrong and I would not defend the CV over the experience of a skilled identifier as they often provide identification steps when asked.
Most often, I use personal knowledge of a species to identify it. For more difficult photos, I may use the CV to double-check if I’m at least in the correct genus. Like previously mentioned, the CV has a hard time noticing the more minor details which may differentiate a species.
CV seems to be pretty clever about species groups in regions that have been thoroughly revised by humans. I am impressed by the level of goby species recognition in the Mediterraneis. Still I find the “margin” of error to be intolerable in other groups. Weak pics or multiple species observations complicate the matter. Anyway, CV will eventually be genius. But I think it is a long way still to go for us humans to get CV up to scratch.
I use CV for most of my observations that are plants and insects on Long Island since the NYC area is so well documented on Inat. However, I rely on humans for ID’s over CV always, and if my ID is corrected by someone else with an accurate description I will agree with them, even if it disagrees with the CV.
I do not use CV if the species is found in an area where low Inat observation density is, or if the species is very rare, where I will key it out.
When I sometimes take 400-600 photos per day, it is very difficult to upload all of them by keying them out.
One problem with CV cannot be avoided: at least as far as we are talking about animals, especially insects, many of them are trying to cheat all the time. It is called Müllerian mimicry, and CV is having a hard time with it. Bad-tasting species of unrelated groups converge on the same appearance, and also some not bad-tasting ones bluff . For example: https://www.inaturalist.org/observations/147749908
CV offers Similar Species, which can be useful (not to you, but to me)
I use the AI for batch uploads for “wham bam” quick ID’s for most moths I know. Anything suspicious or unlikely that the AI suggests I will back up to genus or family, so I guess my answer would be I trust the human (myself) to make the call, and the AI only verifies my gestalt choice. At least for moths!
To quote Emo Philips: *A computer once beat me at chess, but it was no match for me at kick boxing". I like this quote because it works on two levels: no computer has the capacity to recognise why this might be funny. Only truly intelligent things can know what fun is.
Your scenario seems very far removed from the current reality. The CV is effective precisely because it does not, and cannot, make informed evaluations of the kind you suggest. It relies 100% on humans for that. What it does instead (roughly speaking) is search through a very large space of possibilities and select the nearest match according to a carefully contrained set of simple criteria. This is exactly how computers manage to “beat” humans at games like chess: they trade off intelligence against time. Computers can perform simple, repetitive tasks much more efficiently than humans, so in a time-controlled situation like a chess match, it’s obvious that they will perform better (given sufficient processing power). If computers were forced to operate at the same “clock-speed” as a human brain, they would have no chance of out-performing a human at even the simplest of tasks.
To bring this back on topic: the only reason for using the CV is as a time-saving measure. It’s good at quickly filtering large sets of possibilities in exactly the same way that a web-search is good at finding the proverbial needle in a haystack of information (thus it should be no surprise to learn that the TensorFlow software that the CV is based on was developed by Google). Of course, I could just go through the haystack straw by straw and find the needle by myself - if I had the time and the inclination. But sometimes life’s too short, so I choose a quick and dirty method like the CV instead. It doesn’t matter if it often only gives me a bunch of a straws that merely look like needles, because that may be enough to suggest how I can find a much more effective search strategy by myself (i.e. by using Real Intelligence).
Having said all that, though, I usually choose not to use the CV, however effective it may sometimes be. I much prefer to take a more scenic route and invest time in learning. It’s just much more fun to do things that way.
I’m ambivalent about the CV. It’s good! But sometimes it’s wrong. If I’m trying to ID something I totally don’t know, I’ll usually use the CV identification but back it up to genus. (Like many others, I’ll also use its suggestion just to avoid typing out the name.)
There are some people here whose identification skills I trust more than the CV’s, sometimes more than my own. I’d like to just agree with their ID’s even if I don’t know enough to confirm, but I shouldn’t. But sometimes I do. And of course sometimes ID’s or corrections from these people make me realize something I didn’t consider earlier about IDing the observation.
And then there are the many identifiers whose skills I can’t assess. I trust them about as much as I do the CV’s. I think one reason I try to just let my observations go after I’ve posted them is so that I won’t be tempted to agree with them blindly.
I’d rather use the CV for genus or family than leave something to languish at phylum or unknown. If it gets the genus or family wrong, humans probably would, too. How many difficult taxa have we seen stay at “Needs ID” because different humans have different ideas about it? How is that better than picking a CV suggestion at the same level where it is stuck?
For me, it depends very much on the taxon concerned.
With Australian moths, I find that it is a great help in getting to identifications. It is very good on the common species, with others I will often compare the species it suggests with all those in higher level taxa.
With Australian marine molluscs, the CV is much less useful, often suggesting completely way-out species.
I guess the difference is the number of observations it has available, we need to add more marine molluscs.
I’m sure real AI will be able to recognize jokes and make new ones.
I’m with nyoni-pete. For me, it depends on the taxon and the location and even the life stage. If I upload a decent photo of an adult moth from here in New England, the computer vision is usually right. If I upload a decent photo of a dragonfly exuviae from here, it’ll be way off. I often use the CV to point me in the right direction, but then I’ll go look up that taxon in an actual field guide, or in iNat’s taxa pages.