Don't use computer vision

I think it may be helpful to limit AI suggestions to the genus level or higher, at least for users under a certain number of IDs. That would prevent the problem of classmates getting each other’s IDs to Research Grade just by agreeing with the suggested species, but still narrow it down enough to attract specialists.

I also agree that accepted range or nearby observations should be weighted heavier. In my experience if the AI is on the right track the “seen nearby” option is usually the correct one, but a lot of times it’s several entries down the list. Maybe “seen nearby” could be made a separate category in the dropdown, similar to the way the “we’re pretty sure this is” suggestions are now?

7 Likes

It will make no difference to the suggestions you receive. The algorithm is a perceived visual similarity match. You will get the same suggestions if you have no location, the real location or a totally fake location entered.

The only thing you will get with the location entered is the seen nearby label, which also has some limitations to it both distance and temporal.

3 Likes

I think it would be a bad edition if done for all members, it’s faster to pick common plant or bird species from AI when it knows them, so it’s a big time saver. Though if there will only be limitations it could work.

8 Likes

A large number of us use the computer vision ID as a shortcut to typing out the full scientific name, treating it like an auto-complete prompt when texting. We do this even when we’re 100% certain of the ID already. I don’t see how iNaturalist could distinguish this use from naive uses of the CV suggestion.

24 Likes

yes, but it’s basically uninformative, because so many people (myself included) use the CV suggested prompts as a typing shortcut even when we already know what we’re looking at.

7 Likes

That’s only the case if you type literally nothing and use the dropdown menu; if you type so much as a single letter and then select a dropdown option, it won’t display the tag. I can’t speak for other users, but in 100% of observations for me concerning this kind of usage, I type at least the first couple of letters of the name first. Indeed, in my experience, this is actually slightly faster than just selecting from the AI suggestions outright; selecting outright requires the system to ‘think’ for a second or two, whereas if I start typing a few letters then the options appear immediately.

3 Likes

Well, no, it’s faster to use AI, just need to wait when photos are uploaded.

Oh interesting. I must already be in the habit of typing literally nothing most of the time, then. Thousands and thousands of the identifications I’ve made are decorated with that symbol. The vast majority of those were made according to my educated and informed opinion, not the fact that the CV suggested it. I would hate for all that work to be ignored!

8 Likes

Perhaps it’s my computer then?

If I click into the box (on desktop) and wait for AI to appear, it’s fractionally slower than if I type in a letter, in which case the dropdown menu instantly appears. The difference is probably less than a second, but it adds up for 100s of observations

2 Likes

Are you entering your ID while still in the uploader? Whether I start typing or just click in the box there to get the drop-down menu, I get the symbol attached to my ID. So I don’t think the fact that the symbol is there can be used to determine that the observer was just using the AI without knowing what the organism is already.

1 Like

In the uploader screen still. If I type so much as a single letter, I don’t get the tag symbol on my ID.

as an example, I entered IDs for all of these https://www.inaturalist.org/observations?on=2020-06-12&place_id=any&subview=grid&taxon_id=3&user_id=thebeachcomber&verifiable=anby typing in a few letters and then selecting the dropdown, none of them have the symbol

1 Like

I hope someone figures how why it works differently for different people–I’m curious now.

It works faster when the images are already processed, so if I click on them right after adding to uploder it will take much longer, but when they are done the program almost immediately shows suggestions.

I’ve been using iNat for just a couple of months and I’m finding the AI feature incredibly helpful and a real stimulus to learn more. Although I wouldn’t define myself as expert in anything particular, I have been active in and around nature for my whole life and this has given me a sound enough knowledge base to be sensitive about what I do and don’t know, or what can and can’t be known. I would never dream of taking its suggestions as the gospel truth, but it has pointed me in the right direction many times, particularly with arthropods (my passion) where often even arriving at family level is an arduous task for the non-expert. With a little push in the right direction to orient my search, I have sometimes managed to arrive at genus level with a fair degree of accuracy. And in just a couple of months I’ve learnt a tremendous amount (and had great fun) in the process. My feeling is that the validity of AI as the basis for identifications is not so much a matter of expert vs. non-expert, as the user’s degree of “taxonomic sensitivity”. Perhaps many of the errors caused by incautious use of AI could be avoided by providing a beginner’s tutorial explaining its strong and weak points and how best to use it as a guide to identification and learning tool, with the aim above all of improving the user’s sensitivity to the possibility of error.

12 Likes

This part of the discussion (what happens if you type a few letters, or don’t) interests me, too. When I am on the upload page, I almost always click on the species and “give the computer a chance to guess” even if I know absolutely what the species is; then if the top suggestion is right I click on it. Doing this gives me an idea of how accurate the Computer Vision is. Or at least I thought it did. However because I export my photos from Apple Photos after adding a “description” that includes the species name, I now realize that the Computer may be reading my suggestion and acting on it – how circular is that!

3 Likes

Is this flag/indicator also returned in exports? Is it possible to search by this?

I had never that symbol. I just looked back over my recent Observations; most of them have this flag. So, it’s not just the FIRST pick. It’s any pick from the list. The ones that don’t have it are where I disagreed and typed something else, usually a higher taxon, because iNat’s suggestions were so inconsistent.

As far as I’m aware, no (but maybe possible somehow with the API??)

That’s correct, selecting any of the suggested AI options gives the tag

1 Like

No, I am afraid that you are not correct. If your OB gets Research grade,it goes to GBIF - international database, which is used in research. So, when careless and incorrect IDed rrecords are added, researchers, who use GBIF have to do lots of clean up and this results in decreasing credibility of GBIF, iNaturalist and citizen science as a whole. iNaturalist is not only about posting your observations, it is also about collecting biodiversity data. Even if it were only kind of observation diary - why should you want incorrect names even in your diary? As to ‘scolding’ from an expert, first thing, think of what I wrote above and second (a bit rough, I apologize beforehand) - please, do not become wilting lillies when you receive a stronger word from an expert. They do not mean to discourage you, they may just be frustrated by many unheeding, unresponsive or just plainly foolishly behaving users. We all love iNaturalist and want it and the data from it to be better.

9 Likes

My suggestion would be to add kind of warning banner on the tutorial or somewhere that adding species-level ID from AI suggestion should be very carefully considered before doing that.

7 Likes

Yes, you do, as long as you abide by iNaturalist guidelines.

Correct. I think it is stated prominently that it is for connecting people with nature.

True. The members of this community are used to train the AI. There could be better on-boarding so that new users would understand its limitations, or the AI could be designed to be limited to more general categories instead of species and genus. So it’s understandable that new users don’t realize that it is far from perfect.

I can, too, but that is no excuse for rudeness or being abrupt. We are a community here and everyone should be polite, no matter how important they are.

First, no one is forced to go through and add identifications, and anyone who has been using iNaturalist for any length of time knows that there are problems with the apps (example: many people apparently do not know how to find comments addressed to them), and that the AI is being trained from observations made by the users, so, again, there is no excuse for rudeness. The goals of identifiers vary, however. Some want to initiate dialogue, but many don’t for valid reasons, one of which is that there are many, many observers who completely ignore helpful comments for whatever reason. So the best thing to do when you do not get any dialogue with an ID is to add a comment yourself asking for help in understanding the ID. There are a few experts who never respond to anyone’s comments, but don’t let that stop you from asking because other people may come across your observation and read your question and answer it. I especially would advise you to make a polite comment whenever anyone adds an identification of an organism in your photo that is other than your intended subject. The subject of your observation is your choice, and yours alone, assuming that your subject is in the photo. If they refuse to honor your choice of intended subject (fortunately, those identifiers are few), you can opt out of Community ID so that the observation has your intended subject (it will not be eligible for Research Grade, however, until you opt back in to Community ID).

Yes, it is. It has also been a source of frustration for some identifiers, including me.

This would help a lot.

12 Likes