AI Generated or Enhanced Images as "observations"

I agree it’s difficult to know where to draw the line. For example, modern, low-end phone photo editing software can do things like adjust exposure, which is usually going to be considered fine for photo processing.

But simple photo editing apps and programs can now also automatically detect the sky using “AI” (a word that’s nearly lost all meaning at this point, haha), separating it from the foreground/subject. Thus you could lighten the subject of the photo without turning the sky blindingly white so that the image is hard to see. Is this use of AI too much or just fine? Would it be fine if you went in and traced around the subject and did it manually?

What people mostly object to is generative AI but there’s no real hard line between that and automatic photo processing (and all digital cameras have some level of automatic photo processing).

Obviously wholesale made up images should be against the rules. However, even where a program might fill in the gaps is a bit of a grey area to me. For example, here’s an observation where it worked in my favor, though I’m not sure how acceptable people think it is.

When I take a photo with my phone out the window at night it is literally just a black void on the screen. But when I open the image, the automatic photo processing is able to figure out ridiculous amounts of information from that photo, and automatically processes it into something like this. (It doesn’t even save the black void that the photo was originally.)

https://www.inaturalist.org/observations/244253752

The phone is obviously filling in the blanks, and it has that kind of painterly effect recognizable from AI-generated images. But that is what I saw out the window and the reason I was taking the photo. If I shoot in raw, both my desktop programs and phone apps can pull even more information from the photo.

This is the automatic settings of the phone. If I turned off automatic editing, I wouldn’t have been able to get a photo of the skunk I could see at all. With a bug or a plant, this kind of processing/editing could result in an un-IDable fantasy. Should be unacceptable across the board?

Is it likely to result in false positive IDs for things like bugs and plants, or more likely to just create un-IDable messes?

Enhancing is not new. Enhancing does not fundamentally alter the image. For 30 years, image editing software has had Auto Enhance functions, and I think you would be hard pressed to find a non-raw digital image that has not recieved some form of Auto Enhancement, as this is fundamentally part of how digital cameras process the JPEG format when originally capturing the image. AI image enhancement is is no different than Auto Enhancement, unless it is recreating the image and adding additional details that were non-existant in the original.

AI enhancement is not the same thing as AI Generation. They are worlds apart.

Oh good. I am glad you get it. Your firm grip on reality will serve you well in this ever changing world. “Unaltered entropy”. I like that. Sleep well my friend. Atonement is tiring work.

They need to be banned, have their account and observations deleted. This is an intentional attempt to contaminate scientific databases so they can be discredited in the future!

At a minimum, the user should be obliged to upload an unaltered original / raw image alongside the “enhanced” image.
Often, the phone will have a raw image stored, and applies the “enhancements” when using / exporting the image.
May be worth looking into the way that media firms are verifying that images are real, and impliment something similar on iNat.
This will become essential, and must be addressed now, not in guidelines, but in the system itself. It cannot re relied on that users will “behave”, or for others users to hopefully notice that something is not real. There is no space for being PC, or polite, it’s reallality / research, not entertainment.

Well said. Thank you. I have also been doing photography for decades. Many of these posts show a lack of understanding of photo processing. “AI images should be banned” - what does that even mean? Bringing the subject out of the shadows, seeking accuracy by adjusting color and tone, removing distracting noise, etc., all improve the quality and value of the image when done correctly. For those who want to take and post only unaltered snapshots, that’s their choice. In many of these snapshots, minimal processing would have made the subject easier to see, thereby enhancing the image’s value. However, I would never suggest that poor-quality snapshots should be banned. Fear of AI as it creeps into our lives is understandable, but know what you’re talking about before suggesting “banning” its use.

Upscaling with ai generates pixels from those that are already present to create better resolution. Every pro photographer knows higher pixel sensors usually mean better photos. That’s why we pay the big bucks for digital cameras with quality sensors. No serious photographer would use this technology if it was “making sh*t up” and generating or hallucinating something that wasn’t already there, like wing veins. You only want the photo as taken to be sharper and with higher resolution, in some cases for printing quality.

Every single digital camera not shooting in RAW mode “processes” photos in the camera according to its algorithms. As you get further into photography you tend to take RAW photos and do all that editing yourself, including features a typical camera JPEG would give you. People brag all the time about the quality of smart phone photos. That’s the camera, with its built in ai deciding what to enhance, sharpen, dehaze, lighten, darken, highlight, etc. The next step for many people is continuing this editing in a program like Lightroom.

You certainly can dramatically edit features like colors, cropping, etc in Lr AND generate something in a photo with ai. But you need to direct that to happen, the software doesn’t do this without you telling it to.

That doesn’t change that I’ve encountered (rarely) Chironomids with AI edits or whatever they used that made them unidentifible becuase they had wing veins that were essentially made up. Every wing vein on a Chironomid has a name.

The above was speaking from experience that.

So, you are not firmly bound to the “raw is law” tribe and culture? :rofl:
Lucky you!
Seems they are already now an endangered species, hardly protectable by photo-evidence only. Maybe some will survive in isolated network enclosures!
I am so happy to see people discussing the difference between an “analog adaequate” photo of a butterfly, an AI ammended photo and an AI generated picture of a butterfly in relation to evidence and data quality. Very soon, the human eye will no longer be able to make a proper distinction.
Without legal obligation to mark AI contribution to data (hopeless idea…) and if INat has the desire to set standards to limit such Artificial Information, i do not see how this should be done without advanced image analysis, maybe involving AI. Better clash of AI algorythms than clash of cultures?

And expecting iNaturalist users to do that would be just one more form of elitism.

Well, not yet. But the apocalypse is looming…

Could there just be a tutorial in Inat for all new users? Explaining the rules and stuff. I feel that this would reduce the Ai, duplicate and fake obs by a huge amount.

| Tutorials for using iNaturalist | : https://help.inaturalist.org/en/support/solutions/151000049022

| What kind of photos or sounds should I attach to observations? | : https://help.inaturalist.org/en/support/solutions/articles/151000169928-what-kind-of-photos-or-sounds-should-i-attach-to-observations-


There’s a balance between providing new users with enough information to use the site productively (however one wants to define that) on the one hand and making the registration process so onerous that users give up before finishing it on the other.

Raw images are very large files and not supported currently by iNat. All other formats do have some kind of automatic photo processing done to them.

My phone does not save raw files unless I shoot using a specific app (which is memory and battery intensive). My camera also has jpg or raw (NEF) settings. Raw is preferable in terms of photography from so many standpoints, but it means carrying a lot of extra batteries and SD cards, which can be a pain for multi-day camping trips. And it takes a very long time to edit raw files manually. Worth it for great wildlife shots I want to use for a contest or make into holiday cards, but not really something I can motivate myself to do for 900 photos of every organism I saw intended for iNat. It’s a bit sacrilegious from a photographer’s standpoint, but I usually set my camera to just JPG for everyday outings.

I hope that there is a clear distinction for AI enhancements, such as noise reduction. It is baked into many modern camera firmware, so it might not be possible to disable on some systems, and also available in most editing software. Especially for night, or dusk photography, like fruit bats it is very useful and hopefully acceptable. AFAIK it doesn’t generate anything that isn’t already in the photo, but “cleans” the noise.

My impression reading this and other similar threads is that the term “AI” has become something of a bogeyman. Anything that involves a technology which could even remotely be associated with so-called Artificial Intelligence tends to be lumped together and dismissed automatically as detrimental. It’s not that simple. Considering just the field of photography and iNaturalist, of course observations in which the image is wholly or partly generated by AI should be strictly banned. It goes without saying. But you have to bear in mind that no photograph is actually reality. All images shot in jpg have already undergone some form of automatic in-camera (or in-smartphone) processing before you even see them, you have little choice in the matter. And even with raw, the choice of camera and lens can radically change an image, without any form of post-processing.

In my view, when documenting an organism (as opposed to artistic representations), the aim of both the act of photographing itself and all subsequent post-processing should be to bring that image back as close as possible to reality as perceived by the photographer. In the case, of raw files in particular, post-processing is essential in this and intelligent use of AI based tools can play an important part. The difference between “good” and “bad” use of these tools has little to do with artificial intelligence and everything to do with human intelligence. And actually I worry a lot more about that than I do about AI.

We need to find out which tool or feature adds surplus details to photos so the guidelines can assist observers avoiding this pitfall.

I doubt upscaling or noise reduction would do that. I guess it is the “creative” features of phones or editing tools to blame but some photo expert advice is needed.

If I kept on with landscape photography only I would be using RAW by now.
It is not an option now. A grass that is new to me needs 10-12 clear photos for me to identify it, that is about 30 photos to choose from. I usually have over 300 photos a day I am out. Even sorting and basic edits take a lot of time.

That’s why I would like to go with a more basic “accuracy” Data Quality Assessment. One could also not use AI but use “traditional” ways of removing or adding things to the photo, like cloning and stamping. The point is that the photo (or sound) is evidence that is being submitted for evaluation.

I think the worry with AI-generated stuff is that it’s so much easier for people to make serious alterations to a photo that may be more difficult to detect.

Thanks for this! I have a recent Moto with updated Android OS and had no idea these settings existed. I use the macro setting for arthropods etc. and I’m fairly appalled at the idea that the resulting photos have been “enhanced” without my consent.

ETA: and yup, there it was. For heaven’s sake.