The difference is that in a drawing, all the features of the drawing were intentionally added by the human, so they represent what the human saw – or at least, what they thought they saw. Of course there is some possibility for the inexperienced observer to introduce error at this point (e.g., if they saw something only fleetingly but are so convinced it is a particular thing that they add details that they expect to be there but didn’t see). In this context, think it is reasonable to evaluate the plausibility of the documentation, but this is not fundamentally different than considering supplementary evidence in notes provided by the observer (“tastes bitter”, “smelled like mint”, “heard tee-wit call” etc.).
However, with generative AI, whether it is drawing something from scratch or enhancing a photograph or drawing, all the details that it adds are its own creation, based on some model or another in its training set that may or may not have any relation to what the observer saw. Now, theoretically the observer would be carefully examining these AI images, but most of us don’t have photographic memories, so we won’t be able to be certain that every single detail in the images corresponds with the individual we saw. And these “made up” details might be diagnostically relevant.
The other problem with generative AI is that its images do not at first glance look like artificially created images – unlike a drawing, which is immediately recognizable as a drawing. This is a problem both for humans looking at observations and for the CV, because AI-generated images might be treated as though they were photographs. I see drawings as unproblematic here precisely because they are not going to be confused with photographs. This means that humans know they are seeing a field drawing, with all the caution that needs to go into evaluating such evidence. It also means that the CV will treat such images as outliers – as a different “pattern” than the photographs of that species. If someone uploads a drawing to the CV, it will look for similar images in its training set, which will likely be other drawings. But I doubt that people uploading field drawings expect to be able to rely on the CV to provide a correct guess anyway.
I see what you’re saying, but with AI images you can really just keep generating images an unlimited amount of times until you’ve created one that best resembles the organism that you saw. Although an user might draw an organism, their drawings can be just as poor portrayals of an organism anatomy as what an AI is capable of. For instance someone might have seen a 7-spot ladybird, but failed to capture a photo of it. Because the Data Quality Assessment only asks for evidence and doesn’t require it to be specifically a photograph that opens up the door to other forms of representation. A user might be a very poor drawer and fail to capture any details, while they do remember very well what they saw. With that in mind what is stopping someone from using an AI to generate some ladybird images and continue generating images until the features match what they saw?
^ This example was made on my 2nd prompt attempt. It would likely be far better than the drawings most people are humanly capable of. If the user saw a 7-spot ladybird then surely this could be used as evidence as it’s a depiction of what they saw, same as how a drawing would be. The user may not have drawn it by hand, but they created a prompt and continued generating images until they got a close match with what they think they saw, which is similar to how someone would continue drawing until their drawing was a match with what they think they saw.
The problem is, that any AI image will be more detailed than the observer remembers. As @spiphany said, the AI adds stuff that has nothing to do with the original, which is inherently unscientific. It’s the same level of unscientific as just inventing a bunch of data so the graphs look nicer. (The integrity of the data is also the reason we are told to draw what we see and not what we expect or know about the species in zool. and bot. art)
For IDing purposes, it may not be a problem for a very IDable species, but for more cryptic species, small details can mean a lot. A drawing without all the necessary details will not be identifiable to species, but, assuming the observer is trustworthy, it is an honest representation of the available data. An AI generated image with invented details, may be identifiable to species …but perhaps to the wrong species.
I think if drawings are going to be accepted they should be very detailed and well done, time stamped and maybe even a picture of the habitat as well to confirm the location and that the person was actually there and actually did see this plant/animal/etc.
Otherwise people will start submitting all kinds of drawings and paintings whether they actually observed the organism or not.
I haven’t posted any drawings as observations.
I am here to state my opinion on this topic
Yes, drawings are evidence.
If drawing what you see in nature is how you best interact with nature, you have indeed fit the first basic co-reason behind iNaturalist.
Too many people are focused on the skill level of the artist. I think they have tried more honestly than the 50mph roadside photos.
I agree as I do not believe using AI is drawing.
I don’t consider using AI as art in any form.
It’s not anymore valid than using someone else’s photo you find and copy.
With drawings, there’s nothing we can do to check whether people really observed the organism, true. But then again, people can also submit random photos found on the internet, and that’s not easy to police. Or they can get the location deliberately wrong, to make an observation look special. Or…? There are lots of ways people can produce incorrect data if they really want - ultimately, I think we have to assume good faith unless there’s good reason to think otherwise (as was said earlier). I don’t think we should be too ready to make decisions based solely on how people might choose to abuse the system.
As far as needing to be well done, drawings that are too poorly done are unlikely to get to RG anyway, so I don’t see why there should be a need to worry.
PS If I did any drawings, they wouldn’t prove anything except that I have zero art skill. And a drawing of the habitat? You might think I was from Mars and had never even seen Earth! :)
But why hold drawings to a higher standard than photos? Many cameras (including mine) don’t add location information, so I have to add them manually. Since my photos often don’t show much of the habitat, I could essentially set the location to who-knows-where. At some point, you will ultimately need to rely on trust.
Regarding image “quality”: As an IDer, I ID a lot of low-res photos which aren’t very detailed to whatever level I’m confident with. In many cases, a simple drawing would probably convey more information than the blurry sphere that may be a beetle, dirt, or anything else, really.
Here is an example of a very simple drawing I included in an observation of a very fast moving ciliate. It’s not a very detailed drawing by any means, but arguably, the photos in that same observation are significantly worse.
As with many discussions, I think we are forgetting the other half of iNat’s aim/ purpose: Connecting people with nature. If you are curious about something you saw, but didn’t get a photo, then IMO, by all means, make a drawing based on what you recall, and post it. No matter how good your drawing skills.
I would say that a drawing is evidence. Like any evidence, you have to evaluate its quality. Some photos are awful, and some drawings are good. Maybe a photo or a drawing is only good enough to say it’s a plant. If that’s the case, then that’s as far as the identification can go.
This thread is so long ago I don’t recall if I participated last time, but:
The issues people have with drawings is that they are much more vulnerable to being false evidence than photos can be. iNaturalist in general works on an evidence-based system, and while it is possible to be malicious or misleading with the submitted data, nothing is quite as easy to fake as a drawing or sketch of what you claim to have seen.
Generally speaking I think only a very small fraction of people can be trusted with posting drawings. And that’s extremely hard to police for many reasons. Personally, I am very wary about drawings as evidence.
But the site rules and staff I believe permit them, so that’s the only answer that matters for iNat observations. There is always the “evidence of organism” data field if something seems very suspect or clearly wrong.
The fact that making a sketch involves a reasonable amount of effort (far more than googling for someone else’s photo) reduces this risk substantially. I’m not sure there’s much to be gained for anyone by submitting lots of faked observations based on drawings.
Not to mention the fact that (as this thread shows) sketches tend to be viewed a lot more critically by IDers – i.e., they are less likely to be ID’d and less likely to become RG. So drawings are probably not an effective or efficient way to fake observations if one is doing so to pretend one saw something one didn’t.
But I think the risk with drawings is not cases of intentional deceit, but where the observer saw something that they are sure is something remarkable, so they draw what they are convinced they saw – even though it is quite likely it was something more ordinary.
I think evaluating drawings involves as much assessing the context and the reliability of the account as the drawing itself (is the observer skilled enough to correctly recognize what they saw, what were the circumstances of the sighting, etc.). I can understand that IDers might feel uncomfortable doing this, as it is fairly different from the usual procedure of evaluating photographic evidence (objective evidence, rather than the subjective impression of the observer). This doesn’t mean that I think drawings shouldn’t be allowed; I think there are lots of reasons why they are valuable. But I wonder whether rather than being concerned about drawings potentially becoming RG and trying to find ways to prevent this from happening, IDers might be better off just skipping such observations if they do not feel they want to or are able to perform this sort of assessment.
I am a bit late to this discussion as I haven’t had access to iNaturalist for a few days. As I scan all the British and Irish Crustacea, I ought to have seen this one when it was first posted, but instead I have read this forum discussion first. If I had gone to the observation first, I don’t think I would have concluded Gammaridae. The margin notes say shrimp-like, which would make me think it isn’t a shrimp. If you meant true shrimp, Crangon, then there isn’t a rocky stream equivalent in the British Isles. If you meant like a freshwater amphipod such as Gammarus, that means you know what an amphipod looks like and implies this had similarities but was different in some way. From the drawing I would have gone for Simulium larva (despite them not reaching 1.5 cm) or a caseless caddis larva.
Regarding the main question, yes drawings can certainly be evidence. Microscopic structures can be much easier to see than to photograph and a sketch can be the vital ingredient to an identification. But people submitting drawings should be open to questioning on how they achieved it and why they opted for a drawing, just as photographers should expect questions if the photos look strange in some way. I don’t see why people submitting drawings would be more likely than photographers to post fictional observations. But with drawings from memory, there could be the temptation to fill in a gap in your recall by checking in a book how something ought to look.
statements like this are so odd to me. Linneaus did not have a camera so i guess all of his taxonomic system is unscientific and i don’t need to worry about lumpers or splitters any more because Linnean taxonomy is now cancelled. And poor Darwin not having an iphone on the Beagle…
I think it depends on the nature of a drawing, but when i do actual botanical sketches i pay MORE attention to detail than i do to most photos, except those i take while dissecting and keying out a plant or something. So while i understand the sentiment i don’t agree with the statement. Of course, i don’t add botanical sketches any more on here, mostly because i don’t want another reason for people to argue with me. It would be neat if it were encouraged but just tracked differently or something. iNat isnt’ the place it used to be, and i don’t know if it’s still the ‘nicest place on the internet’ as someone voted it once (since other sites are even worse, yes) but it sure isn’t as nice as it used to be.
Not to mention that most actual botanical sketches show more details than photos. The trend to use digital photography for everything is so frustrating to me because it means that most wildflower guides are useless; my old Peterson guides show the differences between lookalike species that photo-based guides would dismiss by saying, “there are many similar species.”
And in addition to these functional reasons, I think aesthetic and sentimental reasons are important as well. It’s a pity that science has neglected this side more and more to create some questionable sterile objectivity appearance.
